feat: event-based walker detector tuned to real 7' overhead mount

Replace per-track line-crossing counter with a single event state machine
gated by foreground pixel count (ENTER=250, EXIT=150) and finalized by
quiet-exit or timeout. Direction inferred from centroid excursion
(up_score vs down_score) on quiet-exit fires, and from net displacement
(last_c vs first_c) on timeout fires.

Tuning reflects bench data at the intended 7' overhead mount: walkers
produce smaller centroid excursions than originally modelled, so
EXTENT gates, MIN_TRAJ, MAX_FRAMES and REFRACTORY were all relaxed from
their initial guesses. Constants and rationale live in firmware/lib/cv/cv.h.

Bench results (8 isolated walks, 4 entries + 4 exits):
  * Event detection: 8/8 (100%)
  * Aggregate entries+exits split: 4+4 (matches)
  * Per-walk direction labelling: 4/8 (~50%)

Document explicitly that per-walk direction is unreliable at this mount
and that downstream analytics should trust only gross traffic
(entries + exits). Recovering direction would require a physical mount
change or a richer signal; both are out of scope for v1.

Tooling:
  * tools/replay_logs.py — replay event state machine against captured
    [F] diagnostic lines, for offline tuning without flash-test loops.
  * firmware/src/main_capture.cpp + tools/capture_frames.py +
    tools/replay_frames.py — raw-frame capture firmware and Python port
    of the detector, kept in tree for future iteration even though the
    TimerCamera-F serial driver stripped specific byte ranges in testing
    and log-based replay became the working path.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
This commit is contained in:
2026-04-17 16:03:36 -07:00
parent 3b471992f2
commit a37207b6ff
12 changed files with 1203 additions and 340 deletions

5
.gitignore vendored
View File

@@ -1 +1,6 @@
.worktrees/ .worktrees/
.agent/
.claude/
graphify-out/
firmware/.pio/
*.log

View File

@@ -1,6 +1,8 @@
# DoorCounter # DoorCounter
Retail door traffic counter using M5Stack TimerCamera-F (ESP32 + OV3660). Counts entries/exits via overhead camera CV, passively scans BLE foot traffic, and reports hourly to `logs.research.bike`. Retail door traffic counter using M5Stack TimerCamera-F (ESP32 + OV3660). Counts walker traversals via overhead camera CV, passively scans BLE foot traffic, and reports hourly to `logs.research.bike`.
> **Known limitation — directional accuracy.** This firmware reports counts as `{entries, exits}` for API compatibility, but **per-walk direction labelling is not reliable at the current mount (7' overhead, straight down).** In bench testing, event detection was 100% (8/8 walks detected) while per-walk direction matched the physical walk only ~50% of the time — the centroid trajectories produced by entries and exits were nearly indistinguishable. **The number to trust is gross traffic: `entries + exits` ≈ total walkers through the doorway.** The directional split is an unreliable best-effort heuristic. See [Directional counting](#directional-counting) for why.
## Hardware ## Hardware
@@ -21,7 +23,7 @@ pio run -t upload --upload-port /dev/ttyUSB0
| Module | Behavior | | Module | Behavior |
|--------|----------| |--------|----------|
| CV pipeline | 5 fps, 96×96 grayscale, blob tracking, directional traversal count (origin→destination, once per track) with per-direction cooldown safety net | | CV pipeline | 5 fps, 96×96 grayscale, event-based walker detector (foreground-count state machine; centroid-trajectory direction heuristic) with post-fire refractory period |
| Detection LED | Single blink on entry, double blink on exit (preserves upload/no-WiFi status LED) | | Detection LED | Single blink on entry, double blink on exit (preserves upload/no-WiFi status LED) |
| BLE scanner | Continuous passive scan; deinits during hourly upload to free heap | | BLE scanner | Continuous passive scan; deinits during hourly upload to free heap |
| Reporter | Hourly HMAC-signed POST; 60s boot report for fast connectivity check | | Reporter | Hourly HMAC-signed POST; 60s boot report for fast connectivity check |
@@ -33,31 +35,70 @@ pio run -t upload --upload-port /dev/ttyUSB0
- **First report**: 60 seconds after NTP sync (connectivity check) - **First report**: 60 seconds after NTP sync (connectivity check)
- **Subsequent reports**: every 3600 seconds - **Subsequent reports**: every 3600 seconds
### Directional counting ### Counting model — event-based walker detector
Each tracked blob fires at most **one** event over its lifetime, and only The CV pipeline is a **single event state machine** (no per-blob tracking
when it has genuinely traversed the frame — specifically, when its spawn for counting). Per-frame foreground pixel count gates event start and end;
position and current position are both at least `CV_TRAVERSAL_MARGIN_PX` centroid trajectory within the active event decides direction.
(14 px ≈ 15% of the 96×96 frame) from the line, and on opposite sides.
- Top half → bottom half traversal = **entry** **Event lifecycle:**
- Bottom half → top half traversal = **exit** 1. **Idle → Active**: `fg_count ≥ CV_EVENT_ENTER_THRESH` (250 px) fires event start.
Background updates freeze while the event is active so the walker does
not get absorbed into the baseline.
2. **Active accumulation**: every frame updates `first_c` (once), `min_c`,
`max_c`, `last_c`, `min_y_seen`, `max_y_seen`, and the frame count.
3. **Active → End** (either):
- **Quiet exit**: `fg_count < CV_EVENT_EXIT_THRESH` (150 px) for
`CV_EVENT_QUIET_FRAMES` (3) consecutive frames — walker has left.
- **Timeout**: `event_frame_count > CV_EVENT_MAX_FRAMES` (25 frames ≈ 5s).
4. On end, the event is finalized: gated by minimum duration, vertical
extent (must span a large fraction of the frame), and minimum centroid
trajectory magnitude. Background snaps to the current frame.
5. A **refractory period** (`CV_EVENT_REFRACTORY_FRAMES` = 10 ≈ 2s) after
a fire blocks a new event from starting — absorbs residual lingering
motion that would otherwise double-count.
A blob that appears near the line and wobbles across it does not count **Direction heuristic** (applied only if the event passes all gates):
(both positions are within the margin band). A blob that fully traverses - `up_score = first_c min_c` (how far centroid excursed upward)
then reverses under the same track also does not double-count (the track - `down_score = max_c first_c` (how far it excursed downward)
is flagged `counted`). If tracking churns — the track dies mid-traversal - Quiet-exit events: `is_entry = (up_score ≥ down_score)`
and respawns on the other side — a new track with a new spawn on the - Timeout events: `is_entry = (last_c < first_c)` — net displacement is
crossed side is the normal path to a correct count. more reliable than excursion when the walker is still in frame at timeout.
See `firmware/lib/cv/cv.h` for margin and `cv.cpp` for the crossing logic. Per-mount convention: centroid moving **up through the frame** (y decreasing)
= **entry** into the store.
### Crossing cooldown (safety net) ### Directional counting — known limitation
On top of directional counting, each direction enforces a cooldown between **Per-walk direction labelling is unreliable at the current mount.** In
counted events. Default: `CV_CROSSING_COOLDOWN_FRAMES = 5` (≈0.8s at 5 fps). bench testing (8 alternating entry/exit walks at 4s intervals, 7' overhead
Entries and exits maintain separate cooldowns, so a real entry immediately mount pointing straight down):
followed by a real exit still counts both.
- **Event detection**: 8/8 (100%) — every walk produced exactly one event.
- **Aggregate split**: 4 entries + 4 exits — matches the 4+4 ground truth.
- **Per-walk direction**: 4/8 (50%) — essentially a coin flip.
At this mount, entries and exits produce nearly identical centroid
trajectories: both begin near mid-frame (walker is already large when
`fg_count` crosses 250), both reach a peak excursion toward the top, and
both end near mid-frame (walker's tail is still visible when `fg_count`
drops below 150). No heuristic over the recorded centroid statistics
separates them with better than ~50% accuracy on alternating walks.
**What we ship, and what the server should trust:**
- **Gross traffic (`entries + exits`) is accurate.** This is the number
downstream analytics should use as "people through the door this hour."
- **Directional split is reported but unreliable.** Treat individual
`entries` and `exits` values as a best-effort labelling. Do not infer
net flow or dwell from them.
To actually recover per-walk direction would require either a physical
change (raise or tilt the camera so walkers enter/leave through the frame
edges) or a richer signal than centroid statistics (e.g. time-resolved
optical flow, or a second sensor). That work is out of scope for v1.
See `firmware/lib/cv/cv.h` for tuning constants and `cv.cpp` for the
finalize logic.
## Operator Setup ## Operator Setup
@@ -124,7 +165,7 @@ DoorCounter/
├── firmware/ ├── firmware/
│ ├── platformio.ini │ ├── platformio.ini
│ ├── lib/ │ ├── lib/
│ │ ├── cv/ — CV pipeline (blob tracking, line cross, cooldown) │ │ ├── cv/ — CV pipeline (event state machine, centroid-trajectory direction)
│ │ └── hmac/ — HMAC-SHA256 signing library │ │ └── hmac/ — HMAC-SHA256 signing library
│ └── src/ │ └── src/
│ ├── main.cpp — FreeRTOS tasks, boot sequence │ ├── main.cpp — FreeRTOS tasks, boot sequence

View File

@@ -13,7 +13,7 @@
[TimerCamera-F Device] [TimerCamera-F Device]
├── Provisioning module — captive portal AP on first boot ├── Provisioning module — captive portal AP on first boot
├── Config store — NVS: device_id, location_id, HMAC secret, WiFi creds, line_offset ├── Config store — NVS: device_id, location_id, HMAC secret, WiFi creds, line_offset
├── Camera + CV module — captures frames, runs line-crossing counter ├── Camera + CV module — captures frames, runs event-based walker detector
├── BLE scanner — continuous passive scan (WiFi coexistence mode) ├── BLE scanner — continuous passive scan (WiFi coexistence mode)
├── Report buffer — accumulates counts in RAM, flushes hourly ├── Report buffer — accumulates counts in RAM, flushes hourly
└── HTTP client — HMAC-signed POSTs to logs.research.bike └── HTTP client — HMAC-signed POSTs to logs.research.bike
@@ -89,22 +89,38 @@ Capture → Grayscale → Downscale 96×96 → Frame diff → Threshold → Blob
| Downscale | Bilinear to 96×96 (~11× compute reduction) | | Downscale | Bilinear to 96×96 (~11× compute reduction) |
| Frame diff | Absolute difference against rolling background (updated every ~2s when no motion) | | Frame diff | Absolute difference against rolling background (updated every ~2s when no motion) |
| Threshold | Pixels > 30 intensity delta = foreground | | Threshold | Pixels > 30 intensity delta = foreground |
| Blob detect | Connected components; blobs < 8×8 px discarded as noise | | Event state machine | Single global state machine (not per-blob). Per-frame `fg_count` (total foreground pixels) gates event start and end. |
| Centroid track | Nearest-centroid matching frame-to-frame (max 15px), tracks persist up to 10 missed frames | | Event start | `fg_count ≥ CV_EVENT_ENTER_THRESH` (250 px) → event becomes active. Background updates freeze for the event's duration so the walker does not blend into the baseline. |
| Line crossing | Virtual horizontal line at configurable vertical position (default: 50% of frame height) | | Event accumulation | Each frame records `first_c` (centroid_y at start), running `min_c` / `max_c` / `last_c`, vertical extents (`min_y_seen`, `max_y_seen`), and frame count. |
| Directional traversal | Each track records its **spawn y** and fires **at most once**. An event fires only when the track's spawn position and current position are both ≥ `CV_TRAVERSAL_MARGIN_PX` (14 px) from the line and on opposite sides — i.e. a true traversal, not a wobble. | | Event end | Either **quiet exit** (`fg_count < CV_EVENT_EXIT_THRESH` (150 px) for `CV_EVENT_QUIET_FRAMES` (3) consecutive frames) or **timeout** (`event_frame_count > CV_EVENT_MAX_FRAMES` (25)). On end, background snaps to the current frame. |
| Cooldown | Per-direction cooldown between counted events (default 5 frames ≈ 0.8s @ 5 fps) — safety net on top of directional logic | | Fire gates | Duration ≥ `CV_EVENT_MIN_FRAMES` (5), `min_y_seen ≤ CV_EVENT_EXTENT_TOP` (25) AND `max_y_seen ≥ CV_EVENT_EXTENT_BOT` (50) — event must span a large fraction of the frame — AND `max(up_score, down_score) ≥ CV_EVENT_MIN_TRAJ` (5) |
| Refractory | `CV_EVENT_REFRACTORY_FRAMES` (10 ≈ 2s) after a fire, the machine refuses to start a new event — absorbs lingering motion of the just-counted walker. |
**Counting logic:** **Direction heuristic (applied after fire gates pass):**
- Each track has a `spawn_y` (recorded at creation) and a `counted` flag. - `up_score = first_c min_c` (peak upward centroid excursion)
- An event fires only if the track is **not yet counted**, spawned **firm** on one side of the line (|spawn_y line_y| > `CV_TRAVERSAL_MARGIN_PX`), and is **now firm** on the opposite side. On fire, the track is flagged counted — it will not produce another event for its lifetime. - `down_score = max_c first_c` (peak downward centroid excursion)
- Spawn firm above + now firm below = **entry** - **Quiet-exit fires**: `is_entry = (up_score ≥ down_score)`
- Spawn firm below + now firm above = **exit** - **Timeout fires**: `is_entry = (last_c < first_c)` — walker is still in frame at timeout, so net displacement is a better signal than excursion.
- Cooldown (per-direction, independent entries/exits) is a secondary gate: within `CV_CROSSING_COOLDOWN_FRAMES` of the last counted event in that direction, a new event is suppressed even if a different track's traversal would otherwise qualify.
**Rationale**: a single person traversing the doorway produces one track with a clear origin and destination — that's one count. Shadows that appear near the line and wobble, or tracks that churn at spawn, lack a firm origin and never count. Per-mount convention: centroid moving **up through the frame** (y decreasing) = **entry** into the store.
Counts accumulate as `{entries, exits}` in RAM and reset each hour on report. **Counting surface**: `{entries, exits}` accumulate in RAM and reset each hour on report.
**Directional accuracy is best-effort, not guaranteed.** In bench testing at the intended 7' overhead straight-down mount:
| Metric | Result |
|--------|--------|
| Event detection | 8/8 walks (100%) |
| Aggregate entry/exit split | 4+4 vs ground-truth 4+4 (matches) |
| Per-walk direction labelling | 4/8 (50%) — no better than chance |
At this mount, entries and exits produce nearly identical centroid trajectories: the walker is already large when `fg_count` crosses 250 (so `first_c` is always near mid-frame), their tail is still visible when `fg_count` drops below 150 (so `last_c` is always near mid-frame), and the excursion in between peaks upward for both directions. No statistic computable from (`first_c`, `min_c`, `max_c`, `last_c`, duration) separates them reliably.
**Contract with downstream consumers (API and analytics):**
- **`entries + exits` is the trustworthy number** — it is the count of walkers through the doorway in the hour. Use this as "foot traffic."
- **Individual `entries` and `exits` are reported for API shape compatibility, but should not be relied on for net flow, dwell, or any per-direction analysis.**
Recovering true direction requires either a physical change (tilt or raise the camera so walkers pass fully through the frame edges) or a richer signal (time-resolved centroid trajectory, optical flow, secondary sensor). Both are out of scope for v1.
--- ---

View File

@@ -5,8 +5,21 @@
#include <algorithm> #include <algorithm>
#include <vector> #include <vector>
static void event_reset(CVState& s) {
s.event_active = false;
s.event_start_frame = 0;
s.event_frame_count = 0;
s.event_peak_n = 0;
s.event_first_c = -1.0f;
s.event_last_c = -1.0f;
s.event_min_c = (float)CV_H;
s.event_max_c = -1.0f;
s.event_min_y_seen = CV_H;
s.event_max_y_seen = -1;
s.event_quiet_count = 0;
}
void cv_init(CVState& state) { void cv_init(CVState& state) {
// Initialize members directly — avoid CVState{} temporary which puts 9KB on stack
memset(state.background, 0, sizeof(state.background)); memset(state.background, 0, sizeof(state.background));
state.bg_valid = false; state.bg_valid = false;
state.last_motion_frame = 0; state.last_motion_frame = 0;
@@ -15,8 +28,8 @@ void cv_init(CVState& state) {
state.tracks.clear(); state.tracks.clear();
state.entries = 0; state.entries = 0;
state.exits = 0; state.exits = 0;
state.last_entry_frame = 0; state.last_fire_frame = 0;
state.last_exit_frame = 0; event_reset(state);
} }
void cv_reset_counts(CVState& state) { void cv_reset_counts(CVState& state) {
@@ -26,9 +39,6 @@ void cv_reset_counts(CVState& state) {
struct Point { int x, y; }; struct Point { int x, y; };
// Note: queue may grow to CV_PIXELS entries (~72KB) on large blobs.
// Requires PSRAM (enabled via -DBOARD_HAS_PSRAM in platformio.ini).
// BFS flood fill. Marks visited pixels (sets fg to 0). Returns {-1,-1} if blob < CV_MIN_BLOB_PX.
static std::pair<float,float> extract_blob(uint8_t* fg, int start_x, int start_y) { static std::pair<float,float> extract_blob(uint8_t* fg, int start_x, int start_y) {
std::vector<Point> queue; std::vector<Point> queue;
queue.reserve(512); queue.reserve(512);
@@ -60,7 +70,7 @@ static std::pair<float,float> extract_blob(uint8_t* fg, int start_x, int start_y
static std::vector<std::pair<float,float>> find_centroids(const uint8_t* fg) { static std::vector<std::pair<float,float>> find_centroids(const uint8_t* fg) {
std::vector<std::pair<float,float>> result; std::vector<std::pair<float,float>> result;
static uint8_t fg_copy[CV_PIXELS]; // static to avoid 9KB stack allocation static uint8_t fg_copy[CV_PIXELS];
memcpy(fg_copy, fg, CV_PIXELS); memcpy(fg_copy, fg, CV_PIXELS);
for (int y = 0; y < CV_H; y++) { for (int y = 0; y < CV_H; y++) {
@@ -82,8 +92,62 @@ static void frame_diff(const uint8_t* frame, const uint8_t* bg,
} }
} }
CVResult cv_process(CVState& state, const uint8_t* frame, uint8_t line_pct) { // Decide whether the just-ended event should fire and in which direction.
CVResult result = {0, 0}; // Up-through-frame (centroid excursion from high y toward low y) maps to
// ENTRY per mount convention.
static void finalize_event(CVState& s, CVResult& result) {
if (s.event_frame_count < CV_EVENT_MIN_FRAMES) return;
// Note: no MAX_FRAMES rejection here. An event that runs the full duration
// may still be a valid walker whose fg_count stayed above EXIT_THRESH due
// to a stale bg or an AEC-driven lighting shift. Extent + MIN_TRAJ gates
// below already reject stationary-person / wobble events.
if (s.event_min_y_seen > CV_EVENT_EXTENT_TOP) return;
if (s.event_max_y_seen < CV_EVENT_EXTENT_BOT) return;
// Direction from centroid excursion relative to event start.
// up_score: how far centroid excursed upward (smaller y) from first_c.
// down_score: how far it excursed downward (larger y) from first_c.
float up_score = s.event_first_c - s.event_min_c;
float down_score = s.event_max_c - s.event_first_c;
float winning = (up_score >= down_score) ? up_score : down_score;
if (winning < CV_EVENT_MIN_TRAJ) return;
// Timeout-aware direction. Quiet-exit events (fg fell below EXIT_THRESH)
// have walker fully out of frame → min/max excursion bracket the true
// traversal and up/down scores are reliable. Timeout events (event hit
// MAX_FRAMES while still elevated) captured both an approach and a
// departure within the window, so excursion measures the walker's
// *range in frame* rather than direction — an entry walker who paused
// near the top, then drifted back toward the middle before timeout
// gets (wrongly) called an entry by up-score even though net motion is
// mixed. For those, the net first→last centroid displacement is a
// better direction signal (it's where the walker ended up, not just
// where they peaked).
bool timed_out = (s.event_frame_count > CV_EVENT_MAX_FRAMES);
bool is_entry;
if (timed_out) {
is_entry = (s.event_last_c < s.event_first_c);
} else {
is_entry = (up_score >= down_score);
}
if (is_entry) {
s.entries++;
result.entries_delta++;
} else {
s.exits++;
result.exits_delta++;
}
s.last_fire_frame = s.frame_index;
result.fire_first_c = s.event_first_c;
result.fire_min_c = s.event_min_c;
result.fire_max_c = s.event_max_c;
result.fire_last_c = s.event_last_c;
result.fire_duration = s.event_frame_count;
}
CVResult cv_process(CVState& state, const uint8_t* frame, uint8_t /*line_pct*/) {
CVResult result = {0, 0, 0, -1, -1, -1.0f, 0.0f, 0.0f, 0.0f, 0.0f, 0};
state.frame_index++; state.frame_index++;
if (!state.bg_valid) { if (!state.bg_valid) {
@@ -92,105 +156,147 @@ CVResult cv_process(CVState& state, const uint8_t* frame, uint8_t line_pct) {
return result; return result;
} }
static uint8_t fg[CV_PIXELS]; // static: avoids 9KB on task stack static uint8_t fg[CV_PIXELS];
frame_diff(frame, state.background, fg, CV_PIXELS); frame_diff(frame, state.background, fg, CV_PIXELS);
int fg_count = 0; // Running-average background blend: bg = (31*bg + frame)/32. Adapts to
for (int i = 0; i < CV_PIXELS; i++) fg_count += fg[i]; // slow scene drift during idle periods. Frozen during an active event so
// the walker's signature is never absorbed — otherwise bg retains a
bool motion = fg_count > CV_MIN_BLOB_PX; // "ghost" of the walker for ~30 frames after they leave, keeping fg_count
if (!motion) { // elevated and preventing subsequent walkers from producing a clean
if (state.frame_index - state.last_motion_frame > 10) { // trajectory.
memcpy(state.background, frame, CV_PIXELS); if (!state.event_active) {
for (int i = 0; i < CV_PIXELS; i++) {
state.background[i] = (uint8_t)(((uint16_t)state.background[i] * 31 + frame[i]) >> 5);
} }
}
int fg_count = 0;
int min_y = CV_H, max_y = -1;
long sum_y = 0;
for (int y = 0; y < CV_H; y++) {
const uint8_t* row = &fg[y * CV_W];
int row_count = 0;
for (int x = 0; x < CV_W; x++) row_count += row[x];
if (row_count > 0) {
if (y < min_y) min_y = y;
if (y > max_y) max_y = y;
sum_y += (long)row_count * y;
fg_count += row_count;
}
}
result.fg_count = fg_count;
result.fg_min_y = (fg_count > 0) ? min_y : -1;
result.fg_max_y = (fg_count > 0) ? max_y : -1;
result.fg_centroid_y = (fg_count > 0) ? ((float)sum_y / fg_count) : -1.0f;
// Hard self-heal: if more than half the frame is fg, bg is catastrophically
// wrong. Snap and skip the event machine this frame.
if (fg_count > CV_PIXELS / 2) {
memcpy(state.background, frame, CV_PIXELS);
state.last_motion_frame = state.frame_index;
if (state.event_active) event_reset(state);
return result;
}
// Diagnostic track management (no effect on counting).
bool motion = fg_count > CV_MIN_BLOB_PX;
if (motion) {
state.last_motion_frame = state.frame_index;
auto centroids = find_centroids(fg);
std::vector<bool> centroid_matched(centroids.size(), false);
for (auto& track : state.tracks) {
float best_dist = CV_MAX_MOVE * CV_MAX_MOVE;
int best_idx = -1;
for (int i = 0; i < (int)centroids.size(); i++) {
if (centroid_matched[i]) continue;
float dx = centroids[i].first - track.x;
float dy = centroids[i].second - track.y;
float d2 = dx*dx + dy*dy;
if (d2 < best_dist) { best_dist = d2; best_idx = i; }
}
if (best_idx >= 0) {
centroid_matched[best_idx] = true;
track.x = centroids[best_idx].first;
track.y = centroids[best_idx].second;
track.missed = 0;
} else {
track.missed++;
}
}
state.tracks.erase(
std::remove_if(state.tracks.begin(), state.tracks.end(),
[](const CVTrack& t){ return t.missed > CV_MAX_MISSED; }),
state.tracks.end());
for (int i = 0; i < (int)centroids.size(); i++) {
if (centroid_matched[i]) continue;
CVTrack t;
t.id = state.next_id++;
t.x = centroids[i].first;
t.y = centroids[i].second;
t.spawn_y = t.y;
t.missed = 0;
state.tracks.push_back(t);
}
} else {
for (auto& t : state.tracks) t.missed++; for (auto& t : state.tracks) t.missed++;
state.tracks.erase( state.tracks.erase(
std::remove_if(state.tracks.begin(), state.tracks.end(), std::remove_if(state.tracks.begin(), state.tracks.end(),
[](const CVTrack& t){ return t.missed > CV_MAX_MISSED; }), [](const CVTrack& t){ return t.missed > CV_MAX_MISSED; }),
state.tracks.end()); state.tracks.end());
return result;
} }
state.last_motion_frame = state.frame_index; // Event state machine. Refractory period after a fire blocks new events
// for CV_EVENT_REFRACTORY_FRAMES frames — absorbs lingering-walker motion
// that would otherwise re-trigger a second count.
bool in_refractory = state.last_fire_frame != 0 &&
(state.frame_index - state.last_fire_frame) < CV_EVENT_REFRACTORY_FRAMES;
auto centroids = find_centroids(fg); if (!state.event_active) {
if (!in_refractory && fg_count >= CV_EVENT_ENTER_THRESH) {
std::vector<bool> centroid_matched(centroids.size(), false); state.event_active = true;
state.event_start_frame = state.frame_index;
for (auto& track : state.tracks) { state.event_frame_count = 1;
float best_dist = CV_MAX_MOVE * CV_MAX_MOVE; state.event_peak_n = fg_count;
int best_idx = -1; state.event_first_c = result.fg_centroid_y;
state.event_last_c = result.fg_centroid_y;
for (int i = 0; i < (int)centroids.size(); i++) { state.event_min_c = result.fg_centroid_y;
if (centroid_matched[i]) continue; state.event_max_c = result.fg_centroid_y;
float dx = centroids[i].first - track.x; state.event_min_y_seen = min_y;
float dy = centroids[i].second - track.y; state.event_max_y_seen = max_y;
float d2 = dx*dx + dy*dy; state.event_quiet_count = 0;
if (d2 < best_dist) { best_dist = d2; best_idx = i; }
} }
} else {
if (best_idx >= 0) { state.event_frame_count++;
centroid_matched[best_idx] = true; if (fg_count > state.event_peak_n) state.event_peak_n = fg_count;
track.x = centroids[best_idx].first; if (fg_count > 0) {
track.y = centroids[best_idx].second; state.event_last_c = result.fg_centroid_y;
track.missed = 0; if (result.fg_centroid_y < state.event_min_c) state.event_min_c = result.fg_centroid_y;
if (result.fg_centroid_y > state.event_max_c) state.event_max_c = result.fg_centroid_y;
if (min_y < state.event_min_y_seen) state.event_min_y_seen = min_y;
if (max_y > state.event_max_y_seen) state.event_max_y_seen = max_y;
}
if (fg_count < CV_EVENT_EXIT_THRESH) {
state.event_quiet_count++;
if (state.event_quiet_count >= CV_EVENT_QUIET_FRAMES) {
finalize_event(state, result);
event_reset(state);
memcpy(state.background, frame, CV_PIXELS);
}
} else { } else {
track.missed++; state.event_quiet_count = 0;
} if (state.event_frame_count > CV_EVENT_MAX_FRAMES) {
} // Timeout end: fg still elevated. Snap bg anyway — in practice
// a stuck-high event means bg is stale (walker has merged
state.tracks.erase( // with stale bg, or AEC shifted). Leaving bg stale permanently
std::remove_if(state.tracks.begin(), state.tracks.end(), // poisons subsequent events. If a walker truly is mid-frame
[](const CVTrack& t){ return t.missed > CV_MAX_MISSED; }), // they'll get absorbed into bg, but that's a rare corner
state.tracks.end()); // beaten by the common case of stale bg chaining events.
finalize_event(state, result);
float line_y = (line_pct / 100.0f) * CV_H; event_reset(state);
for (int i = 0; i < (int)centroids.size(); i++) { memcpy(state.background, frame, CV_PIXELS);
if (centroid_matched[i]) continue;
CVTrack t;
t.id = state.next_id++;
t.x = centroids[i].first;
t.y = centroids[i].second;
t.spawn_y = t.y;
t.above_line = (t.y < line_y);
t.counted = false;
t.missed = 0;
state.tracks.push_back(t);
}
// Directional crossing check. A track counts at most once, and only if it
// spawned clearly on one side of the line AND is now clearly on the other.
// This rejects blobs that wobble around the line (shadows, body straddling
// the line, track churn at spawn) — only a true traversal fires an event.
for (auto& track : state.tracks) {
if (track.missed > 0) continue; // only check tracks matched this frame
if (track.counted) continue; // one track = one trip
bool spawned_above = track.spawn_y < (line_y - CV_TRAVERSAL_MARGIN_PX);
bool spawned_below = track.spawn_y > (line_y + CV_TRAVERSAL_MARGIN_PX);
bool now_above_firm = track.y < (line_y - CV_TRAVERSAL_MARGIN_PX);
bool now_below_firm = track.y > (line_y + CV_TRAVERSAL_MARGIN_PX);
if (spawned_above && now_below_firm) {
bool in_cooldown = state.last_entry_frame != 0 &&
(state.frame_index - state.last_entry_frame) < CV_CROSSING_COOLDOWN_FRAMES;
if (!in_cooldown) {
state.entries++;
result.entries_delta++;
state.last_entry_frame = state.frame_index;
track.counted = true;
}
} else if (spawned_below && now_above_firm) {
bool in_cooldown = state.last_exit_frame != 0 &&
(state.frame_index - state.last_exit_frame) < CV_CROSSING_COOLDOWN_FRAMES;
if (!in_cooldown) {
state.exits++;
result.exits_delta++;
state.last_exit_frame = state.frame_index;
track.counted = true;
} }
} }
track.above_line = (track.y < line_y);
} }
return result; return result;

View File

@@ -12,24 +12,63 @@ static const int CV_MIN_BLOB_PX = 64;
static const float CV_MAX_MOVE = 15.0f; static const float CV_MAX_MOVE = 15.0f;
static const int CV_MAX_MISSED = 10; static const int CV_MAX_MISSED = 10;
// Directional counting margin: a track only counts if it spawned and is now // Event-based walker detector. Per-frame zone-flip approaches were direction-
// both at least this far from the line (in pixels). Prevents counting blobs // blind at realistic mounts: a walker traversing top-to-bottom and a walker
// that wobble around the line or spawn on top of it. Value chosen at ~15% of // traversing bottom-to-top produced identical zone-dominance sequences
// the 96px frame: 14px ≈ the typical torso half-width overhead. // (geometric artifact of asymmetric zones + body spanning the line). The
static const float CV_TRAVERSAL_MARGIN_PX = 14.0f; // event approach buffers a whole walker event, then decides direction from
// the centroid trajectory: sign(first_centroid_y - peak_centroid_y) > 0 means
// the centroid moved upward through the frame during the event.
//
// Per-mount convention: UP through frame == ENTRY into store. Flip the camera
// mount or invert the mapping in cv_process if the physical install differs.
// Per-direction crossing cooldown. Any same-direction crossing whose frame gap // fg_count thresholds that gate event start/end. Tuned against a real
// is strictly less than this value is dropped. At 5 fps, a value of 5 → ≈0.8s // 8-walk isolated test (see .agent/walk_isolated_8walks.log). Lower than
// suppression window. Purpose: mask track churn (blob briefly drops below // initial guesses because the 7' overhead mount produces smaller centroid
// min_blob_px, track dies & respawns, re-crosses). // excursions than we originally modelled.
static const uint32_t CV_CROSSING_COOLDOWN_FRAMES = 5; static const int CV_EVENT_ENTER_THRESH = 250;
static const int CV_EVENT_EXIT_THRESH = 150;
// Number of consecutive sub-EXIT frames required to end an event.
static const int CV_EVENT_QUIET_FRAMES = 3;
// Min/max event duration in frames. Below min = too brief to be a walker
// (noise burst). Above max = stationary object or stuck detection.
static const int CV_EVENT_MIN_FRAMES = 5;
// MAX bounds the event duration. Too low (15) cut events off while walker
// was still physically in frame — every fire hit dur=MAX+1 and bg snapped
// with a walker-ghost baked in, corrupting the next walk. Too high (40)
// merged multiple walkers. 25 frames (5s) lets a single walker reach the
// quiet-exit path (fg drops below EXIT_THRESH) before timeout, so bg snaps
// on a clean empty frame.
static const int CV_EVENT_MAX_FRAMES = 25;
// Required vertical extent: during the event, fg must have reached near the
// top of the frame (min_y <= TOP) AND near the bottom (max_y >= BOT). At a
// 7' overhead mount real walkers span fg y≈0..70, not 0..95 — the original
// 10/85 gates rejected most real walks. Relaxed to catch them while still
// filtering small local motion that doesn't span the doorway.
static const int CV_EVENT_EXTENT_TOP = 25;
static const int CV_EVENT_EXTENT_BOT = 50;
// Minimum centroid excursion (max of up_score/down_score) for a valid
// trajectory. At overhead mount walker centroid traverses ~15-40 pixels;
// 15 was too aggressive and dropped clean walks. 5 still filters wobble.
static const float CV_EVENT_MIN_TRAJ = 5.0f;
// Refractory period after a fire. Shorter than originally chosen — at 5 fps
// a second walker can arrive within 2s of the first, especially at busy
// doorways. 10 frames = 2s of back-pressure, tuned to match the gap between
// consecutive isolated walks in the test log.
static const uint32_t CV_EVENT_REFRACTORY_FRAMES = 10;
// Diagnostic only: tracks are kept for spawn logging. Counting does NOT
// depend on tracks.
struct CVTrack { struct CVTrack {
int id; int id;
float x, y; float x, y;
float spawn_y; // y at track creation — used for directional counting float spawn_y;
bool above_line;
bool counted; // fires at most once per track (one track = one trip)
int missed; int missed;
}; };
@@ -42,13 +81,36 @@ struct CVState {
std::vector<CVTrack> tracks; std::vector<CVTrack> tracks;
int entries; int entries;
int exits; int exits;
uint32_t last_entry_frame; // 0 = never; frame_index of last counted entry
uint32_t last_exit_frame; // 0 = never; frame_index of last counted exit // Event state machine.
bool event_active;
uint32_t event_start_frame;
int event_frame_count;
int event_peak_n;
float event_first_c;
float event_last_c;
float event_min_c; // min centroid_y observed during event
float event_max_c; // max centroid_y observed during event
int event_min_y_seen;
int event_max_y_seen;
int event_quiet_count;
uint32_t last_fire_frame; // 0 = never; frame of last counted fire
}; };
struct CVResult { struct CVResult {
int entries_delta; int entries_delta;
int exits_delta; int exits_delta;
// Per-frame foreground diagnostics (populated every call).
int fg_count;
int fg_min_y;
int fg_max_y;
float fg_centroid_y;
// Populated only on a fire frame; zeroed otherwise.
float fire_first_c;
float fire_min_c;
float fire_max_c;
float fire_last_c;
int fire_duration;
}; };
void cv_init(CVState& state); void cv_init(CVState& state);

View File

@@ -7,6 +7,7 @@ platform = espressif32@6.6.0
board = m5stack-timer-cam board = m5stack-timer-cam
framework = arduino framework = arduino
board_build.partitions = partitions_4mb_ota.csv board_build.partitions = partitions_4mb_ota.csv
build_src_filter = +<*> -<main_capture.cpp>
build_flags = build_flags =
-DBOARD_HAS_PSRAM -DBOARD_HAS_PSRAM
-mfix-esp32-psram-cache-issue -mfix-esp32-psram-cache-issue
@@ -23,6 +24,25 @@ lib_deps =
h2zero/NimBLE-Arduino@^1.4.2 h2zero/NimBLE-Arduino@^1.4.2
espressif/esp32-camera espressif/esp32-camera
; Frame-capture build. Strips WiFi/BLE/CV/reporter; streams raw 96x96 frames
; over serial at 921600 baud for offline algorithm iteration.
[env:timercam-capture]
platform = espressif32@6.6.0
board = m5stack-timer-cam
framework = arduino
board_build.partitions = partitions_4mb_ota.csv
build_flags =
-DBOARD_HAS_PSRAM
-mfix-esp32-psram-cache-issue
-DCORE_DEBUG_LEVEL=0
-DCONFIG_SPIRAM_USE_MALLOC=1
build_src_filter = -<*> +<main_capture.cpp> +<camera.cpp>
monitor_speed = 460800
upload_speed = 115200
upload_flags = --no-stub
lib_deps =
espressif/esp32-camera
[env:native] [env:native]
platform = native platform = native
test_framework = unity test_framework = unity

View File

@@ -55,12 +55,27 @@ static void check_factory_reset() {
// Camera + CV task — runs on core 1 at 5 fps // Camera + CV task — runs on core 1 at 5 fps
static void task_camera(void*) { static void task_camera(void*) {
static uint8_t frame[CV_PIXELS]; // static: avoids 9KB on task stack static uint8_t frame[CV_PIXELS]; // static: avoids 9KB on task stack
int last_logged_track_id = 0; // diagnostic: log each new track once
while (true) { while (true) {
if (camera_capture_96(frame)) { if (camera_capture_96(frame)) {
if (xSemaphoreTake(s_cv_mutex, pdMS_TO_TICKS(100)) == pdTRUE) { if (xSemaphoreTake(s_cv_mutex, pdMS_TO_TICKS(100)) == pdTRUE) {
CVResult r = cv_process(g_cv, frame, g_cfg.line_offset); CVResult r = cv_process(g_cv, frame, g_cfg.line_offset);
if (r.entries_delta) Serial.printf("[CV] entry +%d (total %d)\n", r.entries_delta, g_cv.entries); for (const auto& t : g_cv.tracks) {
if (r.exits_delta) Serial.printf("[CV] exit +%d (total %d)\n", r.exits_delta, g_cv.exits); if (t.id > last_logged_track_id) {
last_logged_track_id = t.id;
Serial.printf("[CV] spawn id=%d y=%.1f\n", t.id, t.spawn_y);
}
}
if (r.fg_count > 0) {
Serial.printf("[F] n=%d y=%d..%d c=%.1f\n",
r.fg_count, r.fg_min_y, r.fg_max_y, r.fg_centroid_y);
}
if (r.entries_delta) Serial.printf("[CV] entry +%d (total %d) first=%.1f min=%.1f max=%.1f last=%.1f dur=%d\n",
r.entries_delta, g_cv.entries,
r.fire_first_c, r.fire_min_c, r.fire_max_c, r.fire_last_c, r.fire_duration);
if (r.exits_delta) Serial.printf("[CV] exit +%d (total %d) first=%.1f min=%.1f max=%.1f last=%.1f dur=%d\n",
r.exits_delta, g_cv.exits,
r.fire_first_c, r.fire_min_c, r.fire_max_c, r.fire_last_c, r.fire_duration);
xSemaphoreGive(s_cv_mutex); xSemaphoreGive(s_cv_mutex);
if (r.entries_delta) led_blink_pattern(1); if (r.entries_delta) led_blink_pattern(1);
if (r.exits_delta) led_blink_pattern(2); if (r.exits_delta) led_blink_pattern(2);

View File

@@ -0,0 +1,64 @@
// firmware/src/main_capture.cpp
//
// Frame-dump firmware. Replaces main.cpp when building env:timercam-capture.
// Streams raw 96x96 grayscale frames at 5 fps over serial (921600 baud) for
// offline algorithm iteration.
//
// Wire format per frame (little-endian):
// magic uint32 0xDC0FC0DE
// frame_ix uint32 monotonic counter
// millis uint32 ms since boot
// pixels byte[9216] raw grayscale 96x96, row-major
//
// No WiFi, no BLE, no CV. Just camera → serial.
#include <Arduino.h>
#include "camera.h"
#include "cv.h" // for CV_PIXELS
#define LED_PIN 2
#define CAM_FPS 5
#define CAM_INTERVAL_MS (1000 / CAM_FPS)
// Magic chosen from bytes that commonly survive; 'FRM1' ascii.
// Avoid high bytes 0xA0-AF / 0xD0-DF — observed missing from the CH9102 stream.
static const uint32_t FRAME_MAGIC = 0x314D5246; // 'FRM1' little-endian on wire
void setup() {
Serial.begin(460800);
pinMode(LED_PIN, OUTPUT);
digitalWrite(LED_PIN, HIGH);
delay(500);
Serial.println("# capture-mode: 460800 baud, 96x96 gray @ 5fps");
Serial.flush();
if (!camera_init()) {
Serial.println("# FATAL: camera init failed");
while (true) {
digitalWrite(LED_PIN, !digitalRead(LED_PIN));
delay(200);
}
}
digitalWrite(LED_PIN, LOW);
}
void loop() {
static uint8_t frame[CV_PIXELS];
static uint32_t frame_ix = 0;
uint32_t t0 = millis();
if (camera_capture_96(frame)) {
uint32_t ms = millis();
Serial.write((uint8_t*)&FRAME_MAGIC, 4);
Serial.write((uint8_t*)&frame_ix, 4);
Serial.write((uint8_t*)&ms, 4);
Serial.write(frame, CV_PIXELS);
frame_ix++;
digitalWrite(LED_PIN, frame_ix & 1);
}
uint32_t elapsed = millis() - t0;
if (elapsed < CAM_INTERVAL_MS) delay(CAM_INTERVAL_MS - elapsed);
}

View File

@@ -7,258 +7,290 @@ static void fill_frame(uint8_t* f, uint8_t val) {
memset(f, val, CV_PIXELS); memset(f, val, CV_PIXELS);
} }
// Draw a rectangular walker-blob spanning rows [y0, y1], columns [cx-hw, cx+hw].
// Pixel value 200 over background 100 -> frame_diff threshold (30) is cleared.
static void draw_walker(uint8_t* f, int y0, int y1, int cx, int hw) {
fill_frame(f, 100);
for (int y = y0; y <= y1; y++) {
if (y < 0 || y >= CV_H) continue;
for (int x = cx - hw; x <= cx + hw; x++) {
if (x < 0 || x >= CV_W) continue;
f[y * CV_W + x] = 200;
}
}
}
static void prime_bg(CVState& state) {
uint8_t bg[CV_PIXELS];
fill_frame(bg, 100);
cv_process(state, bg, 50);
}
// Let the event state machine see QUIET_FRAMES+1 empty frames so any active
// event finalizes before the next test assertion.
static void quiesce(CVState& state) {
uint8_t bg[CV_PIXELS]; fill_frame(bg, 100);
for (int i = 0; i < CV_EVENT_QUIET_FRAMES + 1; i++) cv_process(state, bg, 50);
}
void setUp(void) {} void setUp(void) {}
void tearDown(void) {} void tearDown(void) {}
void test_frame_diff_no_change_gives_no_fg() { void test_no_change_no_event() {
CVState state; CVState state; cv_init(state);
cv_init(state); uint8_t frame[CV_PIXELS]; fill_frame(frame, 128);
uint8_t frame[CV_PIXELS];
fill_frame(frame, 128);
CVResult r1 = cv_process(state, frame, 50); CVResult r1 = cv_process(state, frame, 50);
TEST_ASSERT_EQUAL_INT(0, r1.entries_delta); TEST_ASSERT_EQUAL_INT(0, r1.entries_delta);
CVResult r2 = cv_process(state, frame, 50); CVResult r2 = cv_process(state, frame, 50);
TEST_ASSERT_EQUAL_INT(0, r2.entries_delta); TEST_ASSERT_EQUAL_INT(0, r2.entries_delta);
TEST_ASSERT_EQUAL_INT(0, r2.exits_delta); TEST_ASSERT_EQUAL_INT(0, r2.exits_delta);
} }
void test_frame_diff_large_change_detected_no_crash() {
CVState state;
cv_init(state);
uint8_t bg[CV_PIXELS], fg_frame[CV_PIXELS];
fill_frame(bg, 100);
fill_frame(fg_frame, 200);
cv_process(state, bg, 50);
CVResult r = cv_process(state, fg_frame, 50);
// Tracking not yet implemented — just verify no crash and result is zero
TEST_ASSERT_EQUAL_INT(0, r.entries_delta);
TEST_ASSERT_EQUAL_INT(0, r.exits_delta);
}
void test_cv_init_clears_state() { void test_cv_init_clears_state() {
CVState state; CVState state;
state.entries = 99; state.exits = 88; state.entries = 99; state.exits = 88; state.event_active = true;
cv_init(state); cv_init(state);
TEST_ASSERT_EQUAL_INT(0, state.entries); TEST_ASSERT_EQUAL_INT(0, state.entries);
TEST_ASSERT_EQUAL_INT(0, state.exits); TEST_ASSERT_EQUAL_INT(0, state.exits);
TEST_ASSERT_FALSE(state.bg_valid); TEST_ASSERT_FALSE(state.bg_valid);
TEST_ASSERT_FALSE(state.event_active);
} }
void test_cv_reset_counts() { void test_cv_reset_counts() {
CVState state; CVState state; cv_init(state);
cv_init(state); state.entries = 5; state.exits = 3;
state.entries = 5;
state.exits = 3;
cv_reset_counts(state); cv_reset_counts(state);
TEST_ASSERT_EQUAL_INT(0, state.entries); TEST_ASSERT_EQUAL_INT(0, state.entries);
TEST_ASSERT_EQUAL_INT(0, state.exits); TEST_ASSERT_EQUAL_INT(0, state.exits);
} }
void test_tracking_spawns_track_for_new_blob() { void test_walker_up_through_frame_is_entry() {
CVState state; // Simulate a walker traversing from bottom to top of frame.
cv_init(state); // Per-frame fg_count and centroid (11-wide column, height H -> n=11*H):
// t0 y=60..95 n=396 c=77 <- event starts (n >= ENTER=300)
// t1 y=30..95 n=726 c=62
// t2 y=0..95 n=1056 c=47
// t3 y=0..60 n=671 c=30
// t4 y=0..25 n=286 c=12 (below EXIT=200, quiet=1)
// t5 y=0..10 n=121 c=5 (below EXIT, quiet=2)
// t6 empty quiet=3 -> finalize
CVState state; cv_init(state);
prime_bg(state);
uint8_t bg[CV_PIXELS]; int rows[][2] = {{60,95},{30,95},{0,95},{0,60},{0,25},{0,10}};
fill_frame(bg, 100); for (int i = 0; i < 6; i++) {
cv_process(state, bg, 50); // init background uint8_t f[CV_PIXELS]; draw_walker(f, rows[i][0], rows[i][1], 48, 5);
// Frame with a bright 30x30 blob in top-left quadrant
uint8_t blob_frame[CV_PIXELS];
fill_frame(blob_frame, 100);
for (int y = 5; y < 35; y++)
for (int x = 5; x < 35; x++)
blob_frame[y * CV_W + x] = 200;
cv_process(state, blob_frame, 50);
TEST_ASSERT_EQUAL_INT(1, (int)state.tracks.size());
TEST_ASSERT_FLOAT_WITHIN(5.0f, 20.0f, state.tracks[0].x);
TEST_ASSERT_FLOAT_WITHIN(5.0f, 20.0f, state.tracks[0].y);
}
static void make_blob_frame(uint8_t* f, int cx, int cy) {
fill_frame(f, 100);
for (int y = cy - 12; y <= cy + 12; y++)
for (int x = cx - 12; x <= cx + 12; x++)
if (y >= 0 && y < CV_H && x >= 0 && x < CV_W)
f[y * CV_W + x] = 200;
}
void test_blob_crossing_line_top_to_bottom_is_entry() {
CVState state;
cv_init(state);
// Line at 50% = y=48, traversal margin = 14px. Spawn must be y<34, final y>62.
// Step ≤14px per frame to stay within CV_MAX_MOVE.
uint8_t bg[CV_PIXELS];
fill_frame(bg, 100);
cv_process(state, bg, 50); // init background
int setup[] = {20, 34, 48, 62}; // spawn firm above, walk across, not yet firm below
for (int i = 0; i < 4; i++) {
uint8_t f[CV_PIXELS]; make_blob_frame(f, 48, setup[i]);
cv_process(state, f, 50); cv_process(state, f, 50);
} }
// Still no count — y=62 is not firm below (needs >62) quiesce(state);
TEST_ASSERT_EQUAL_INT(0, state.entries);
// One more step: y=70 is firm below → entry fires now
uint8_t fcross[CV_PIXELS]; make_blob_frame(fcross, 48, 70);
CVResult r = cv_process(state, fcross, 50);
TEST_ASSERT_EQUAL_INT(1, r.entries_delta);
TEST_ASSERT_EQUAL_INT(0, r.exits_delta);
TEST_ASSERT_EQUAL_INT(1, state.entries); TEST_ASSERT_EQUAL_INT(1, state.entries);
}
void test_blob_crossing_line_bottom_to_top_is_exit() {
CVState state;
cv_init(state);
uint8_t bg[CV_PIXELS]; fill_frame(bg, 100);
cv_process(state, bg, 50);
// Spawn firm below (y=76 > 62), walk toward and across line (y=48), continue
// until firm above (y<34). Each step ≤14px.
int setup[] = {76, 62, 48, 34};
for (int i = 0; i < 4; i++) {
uint8_t f[CV_PIXELS]; make_blob_frame(f, 48, setup[i]);
cv_process(state, f, 50);
}
// y=34 not firm above (needs <34) — no count yet
TEST_ASSERT_EQUAL_INT(0, state.exits); TEST_ASSERT_EQUAL_INT(0, state.exits);
uint8_t fcross[CV_PIXELS]; make_blob_frame(fcross, 48, 22); // firm above
CVResult r = cv_process(state, fcross, 50);
TEST_ASSERT_EQUAL_INT(0, r.entries_delta);
TEST_ASSERT_EQUAL_INT(1, r.exits_delta);
} }
void test_track_spawned_near_line_does_not_count_on_wobble() { void test_walker_down_through_frame_is_exit() {
// Simulates a blob that appears right on the line (e.g. shadow or noise) CVState state; cv_init(state);
// and wobbles across it. With directional margin, no count should fire — prime_bg(state);
// this is the false-positive pattern the feature guards against.
CVState state;
cv_init(state);
uint8_t bg[CV_PIXELS]; fill_frame(bg, 100); int rows[][2] = {{0,35},{0,65},{0,95},{35,95},{70,95},{85,95}};
cv_process(state, bg, 50); for (int i = 0; i < 6; i++) {
uint8_t f[CV_PIXELS]; draw_walker(f, rows[i][0], rows[i][1], 48, 5);
// Spawn within margin of line (y=44, margin=14 so 44 ∈ [34,62])
// then wobble above to y=38, below to y=58. Both within margin.
int setup[] = {44, 38, 58, 42, 56};
for (int i = 0; i < 5; i++) {
uint8_t f[CV_PIXELS]; make_blob_frame(f, 48, setup[i]);
cv_process(state, f, 50); cv_process(state, f, 50);
} }
quiesce(state);
TEST_ASSERT_EQUAL_INT(0, state.entries);
TEST_ASSERT_EQUAL_INT(1, state.exits);
}
void test_approach_retreat_without_full_extent_does_not_fire() {
// Walker approaches from bottom, reaches y=30, retreats, never reaches top.
// Extent gate requires min_y_seen <= 10; this event tops out at y=30 so
// extent never clears and no fire occurs regardless of trajectory score.
CVState state; cv_init(state);
prime_bg(state);
int rows[][2] = {{60,95},{40,95},{30,95},{40,95},{60,95},{80,95}};
for (int i = 0; i < 6; i++) {
uint8_t f[CV_PIXELS]; draw_walker(f, rows[i][0], rows[i][1], 48, 5);
cv_process(state, f, 50);
}
quiesce(state);
TEST_ASSERT_EQUAL_INT(0, state.entries); TEST_ASSERT_EQUAL_INT(0, state.entries);
TEST_ASSERT_EQUAL_INT(0, state.exits); TEST_ASSERT_EQUAL_INT(0, state.exits);
} }
void test_track_counts_at_most_once_even_if_it_wobbles_back() { void test_brief_burst_below_min_duration_does_not_fire() {
// A track that traverses fully should count once. If it then reverses and // One frame of large fg, then gone. Event starts, immediately quiesces,
// crosses back, the track should NOT fire again — it's already counted. // duration ends up below CV_EVENT_MIN_FRAMES.
// (A separate new track on the return trip would count as exit, but while CVState state; cv_init(state);
// the same track persists, it's one trip.) prime_bg(state);
CVState state;
cv_init(state);
uint8_t bg[CV_PIXELS]; fill_frame(bg, 100); uint8_t f[CV_PIXELS]; draw_walker(f, 0, 95, 48, 5);
cv_process(state, bg, 50); cv_process(state, f, 50);
quiesce(state);
// Full traversal top→bottom TEST_ASSERT_EQUAL_INT(0, state.entries);
int walk_down[] = {20, 34, 48, 62, 70};
for (int i = 0; i < 5; i++) {
uint8_t f[CV_PIXELS]; make_blob_frame(f, 48, walk_down[i]);
cv_process(state, f, 50);
}
TEST_ASSERT_EQUAL_INT(1, state.entries);
// Same track reverses back to top. counted=true prevents a second event.
int walk_up[] = {62, 48, 34, 22};
for (int i = 0; i < 4; i++) {
uint8_t f[CV_PIXELS]; make_blob_frame(f, 48, walk_up[i]);
cv_process(state, f, 50);
}
TEST_ASSERT_EQUAL_INT(1, state.entries);
TEST_ASSERT_EQUAL_INT(0, state.exits); TEST_ASSERT_EQUAL_INT(0, state.exits);
} }
void test_cooldown_suppresses_rapid_re_entry() { void test_stationary_large_blob_does_not_fire() {
// Cooldown is a safety net on top of directional counting. Construct two // Static large blob in frame for many frames, then removed. Centroid
// DIFFERENT tracks (each counts once on its own) whose crossings happen // never moves -> MIN_TRAJ gate blocks fire.
// within the cooldown window — the second should still be suppressed. CVState state; cv_init(state);
CVState state; prime_bg(state);
cv_init(state);
state.bg_valid = true;
memset(state.background, 100, CV_PIXELS);
state.frame_index = 100;
state.entries = 1;
state.last_entry_frame = 100;
// Track at y=50 (just below line), spawn_y=20 (firm above) — a valid trajectory. for (int i = 0; i < 10; i++) {
CVTrack t; uint8_t f[CV_PIXELS]; draw_walker(f, 0, 95, 48, 5);
t.id = 1; t.x = 48; t.y = 50; t.spawn_y = 20; cv_process(state, f, 50);
t.above_line = false; t.counted = false; t.missed = 0; }
state.tracks.push_back(t); quiesce(state);
// Frame 101: blob at y=64 (delta=14, matches; firm below line+margin=62). TEST_ASSERT_EQUAL_INT(0, state.entries);
// Would count but cooldown (101-100=1 < 5) suppresses. TEST_ASSERT_EQUAL_INT(0, state.exits);
uint8_t f1[CV_PIXELS]; make_blob_frame(f1, 48, 64); }
CVResult r1 = cv_process(state, f1, 50);
TEST_ASSERT_EQUAL_INT(0, r1.entries_delta); // Wait out the refractory period with bg-only frames so the next walker
// event is accepted.
static void wait_refractory(CVState& state) {
uint8_t bg[CV_PIXELS]; fill_frame(bg, 100);
for (uint32_t i = 0; i < CV_EVENT_REFRACTORY_FRAMES + 2; i++) {
cv_process(state, bg, 50);
}
}
void test_two_sequential_walkers_count_twice() {
CVState state; cv_init(state);
prime_bg(state);
int rows[][2] = {{60,95},{30,95},{0,95},{0,60},{0,25},{0,10}};
for (int i = 0; i < 6; i++) {
uint8_t f[CV_PIXELS]; draw_walker(f, rows[i][0], rows[i][1], 48, 5);
cv_process(state, f, 50);
}
quiesce(state);
wait_refractory(state);
for (int i = 0; i < 6; i++) {
uint8_t f[CV_PIXELS]; draw_walker(f, rows[i][0], rows[i][1], 48, 5);
cv_process(state, f, 50);
}
quiesce(state);
TEST_ASSERT_EQUAL_INT(2, state.entries);
TEST_ASSERT_EQUAL_INT(0, state.exits);
}
void test_full_reversal_counts_entry_then_exit() {
CVState state; cv_init(state);
prime_bg(state);
int up_rows[][2] = {{60,95},{30,95},{0,95},{0,60},{0,25},{0,10}};
int down_rows[][2] = {{0,35},{0,65},{0,95},{35,95},{70,95},{85,95}};
for (int i = 0; i < 6; i++) {
uint8_t f[CV_PIXELS]; draw_walker(f, up_rows[i][0], up_rows[i][1], 48, 5);
cv_process(state, f, 50);
}
quiesce(state);
wait_refractory(state);
for (int i = 0; i < 6; i++) {
uint8_t f[CV_PIXELS]; draw_walker(f, down_rows[i][0], down_rows[i][1], 48, 5);
cv_process(state, f, 50);
}
quiesce(state);
TEST_ASSERT_EQUAL_INT(1, state.entries);
TEST_ASSERT_EQUAL_INT(1, state.exits);
}
void test_refractory_suppresses_back_to_back_fire() {
// After a fire, a second event attempted within CV_EVENT_REFRACTORY_FRAMES
// is suppressed. Simulates walker lingering / ghost re-triggering.
CVState state; cv_init(state);
prime_bg(state);
int rows[][2] = {{60,95},{30,95},{0,95},{0,60},{0,25},{0,10}};
for (int i = 0; i < 6; i++) {
uint8_t f[CV_PIXELS]; draw_walker(f, rows[i][0], rows[i][1], 48, 5);
cv_process(state, f, 50);
}
quiesce(state);
TEST_ASSERT_EQUAL_INT(1, state.entries); TEST_ASSERT_EQUAL_INT(1, state.entries);
// Advance past cooldown; reset a fresh track (previous one had counted=true // Immediate second walker within refractory window — should NOT count.
// set only if it actually counted — cooldown path leaves counted=false so for (int i = 0; i < 6; i++) {
// we reuse the same track). uint8_t f[CV_PIXELS]; draw_walker(f, rows[i][0], rows[i][1], 48, 5);
state.frame_index = 200; cv_process(state, f, 50);
state.tracks[0].y = 50; }
state.tracks[0].spawn_y = 20; quiesce(state);
state.tracks[0].counted = false; TEST_ASSERT_EQUAL_INT(1, state.entries);
state.tracks[0].above_line = false; }
uint8_t f2[CV_PIXELS]; make_blob_frame(f2, 48, 64);
CVResult r2 = cv_process(state, f2, 50); void test_event_counts_after_refractory_expires() {
TEST_ASSERT_EQUAL_INT(1, r2.entries_delta); CVState state; cv_init(state);
prime_bg(state);
int rows[][2] = {{60,95},{30,95},{0,95},{0,60},{0,25},{0,10}};
for (int i = 0; i < 6; i++) {
uint8_t f[CV_PIXELS]; draw_walker(f, rows[i][0], rows[i][1], 48, 5);
cv_process(state, f, 50);
}
quiesce(state);
TEST_ASSERT_EQUAL_INT(1, state.entries);
// Wait out the refractory period.
uint8_t bg[CV_PIXELS]; fill_frame(bg, 100);
for (uint32_t i = 0; i < CV_EVENT_REFRACTORY_FRAMES + 2; i++) {
cv_process(state, bg, 50);
}
// Second walker — should now count.
for (int i = 0; i < 6; i++) {
uint8_t f[CV_PIXELS]; draw_walker(f, rows[i][0], rows[i][1], 48, 5);
cv_process(state, f, 50);
}
quiesce(state);
TEST_ASSERT_EQUAL_INT(2, state.entries); TEST_ASSERT_EQUAL_INT(2, state.entries);
} }
void test_no_crossing_same_side_no_count() { void test_noise_below_enter_thresh_does_not_start_event() {
CVState state; // Tiny 5x5 blob (25 px) never crosses ENTER=300, event never starts.
cv_init(state); CVState state; cv_init(state);
prime_bg(state);
uint8_t bg[CV_PIXELS]; fill_frame(bg, 100); auto small = [](uint8_t* f, int cy) {
cv_process(state, bg, 50); fill_frame(f, 100);
for (int y = cy-2; y <= cy+2; y++)
for (int x = 46; x <= 50; x++)
if (y>=0 && y<CV_H && x>=0 && x<CV_W) f[y*CV_W+x] = 200;
};
for (int cy = 10; cy <= 90; cy += 8) {
uint8_t f[CV_PIXELS]; small(f, cy);
cv_process(state, f, 50);
}
quiesce(state);
uint8_t f1[CV_PIXELS]; make_blob_frame(f1, 48, 20); // above line TEST_ASSERT_EQUAL_INT(0, state.entries);
cv_process(state, f1, 50); TEST_ASSERT_EQUAL_INT(0, state.exits);
uint8_t f2[CV_PIXELS]; make_blob_frame(f2, 48, 30); // still above line, moved closer
CVResult r = cv_process(state, f2, 50);
TEST_ASSERT_EQUAL_INT(0, r.entries_delta);
TEST_ASSERT_EQUAL_INT(0, r.exits_delta);
} }
int main() { int main() {
UNITY_BEGIN(); UNITY_BEGIN();
RUN_TEST(test_frame_diff_no_change_gives_no_fg); RUN_TEST(test_no_change_no_event);
RUN_TEST(test_frame_diff_large_change_detected_no_crash);
RUN_TEST(test_cv_init_clears_state); RUN_TEST(test_cv_init_clears_state);
RUN_TEST(test_cv_reset_counts); RUN_TEST(test_cv_reset_counts);
RUN_TEST(test_tracking_spawns_track_for_new_blob); RUN_TEST(test_walker_up_through_frame_is_entry);
RUN_TEST(test_blob_crossing_line_top_to_bottom_is_entry); RUN_TEST(test_walker_down_through_frame_is_exit);
RUN_TEST(test_blob_crossing_line_bottom_to_top_is_exit); RUN_TEST(test_approach_retreat_without_full_extent_does_not_fire);
RUN_TEST(test_track_spawned_near_line_does_not_count_on_wobble); RUN_TEST(test_brief_burst_below_min_duration_does_not_fire);
RUN_TEST(test_track_counts_at_most_once_even_if_it_wobbles_back); RUN_TEST(test_stationary_large_blob_does_not_fire);
RUN_TEST(test_no_crossing_same_side_no_count); RUN_TEST(test_two_sequential_walkers_count_twice);
RUN_TEST(test_cooldown_suppresses_rapid_re_entry); RUN_TEST(test_full_reversal_counts_entry_then_exit);
RUN_TEST(test_refractory_suppresses_back_to_back_fire);
RUN_TEST(test_event_counts_after_refractory_expires);
RUN_TEST(test_noise_below_enter_thresh_does_not_start_event);
return UNITY_END(); return UNITY_END();
} }

105
tools/capture_frames.py Normal file
View File

@@ -0,0 +1,105 @@
#!/usr/bin/env python3
# tools/capture_frames.py
#
# Read framed 96x96 grayscale frames from the capture-mode firmware over serial
# and write them to a .bin file for offline replay.
#
# Wire format per frame (little-endian):
# magic u32 0xDC0FC0DE
# frame_ix u32
# millis u32
# pixels 9216 bytes
#
# Output file is the raw concatenation of frames (same layout as the wire),
# so replay_frames.py can stream it with identical parsing.
#
# Usage: python tools/capture_frames.py --port /dev/ttyUSB0 --out walk.bin --duration 60
import argparse
import serial
import struct
import sys
import time
MAGIC = 0x314D5246 # 'FRM1' — ascii bytes that survive the CH9102 stream
FRAME_PIXELS = 96 * 96
HEADER_LEN = 12
FRAME_LEN = HEADER_LEN + FRAME_PIXELS
def read_exact(ser, n):
buf = bytearray()
while len(buf) < n:
chunk = ser.read(n - len(buf))
if not chunk:
return None
buf.extend(chunk)
return bytes(buf)
def find_magic(ser):
"""Scan serial byte-by-byte until we see the 4-byte MAGIC."""
window = bytearray()
magic_bytes = struct.pack('<I', MAGIC)
while True:
b = ser.read(1)
if not b:
return False
window.extend(b)
if len(window) > 4:
del window[0]
if bytes(window) == magic_bytes:
return True
def main():
ap = argparse.ArgumentParser()
ap.add_argument('--port', required=True)
ap.add_argument('--baud', type=int, default=460800)
ap.add_argument('--out', required=True)
ap.add_argument('--duration', type=float, default=60.0,
help='Seconds to capture (default 60)')
args = ap.parse_args()
ser = serial.Serial(args.port, args.baud, timeout=1.0)
print(f'# listening on {args.port} @ {args.baud} for {args.duration}s...',
file=sys.stderr)
# Drain boot banner lines.
deadline_banner = time.time() + 2.0
while time.time() < deadline_banner:
line = ser.readline()
if line.startswith(b'#'):
print(line.decode(errors='replace').rstrip(), file=sys.stderr)
if b'capture-mode' in line:
break
deadline = time.time() + args.duration
frames = 0
last_ix = None
dropped = 0
with open(args.out, 'wb') as f:
while time.time() < deadline:
if not find_magic(ser):
continue
body = read_exact(ser, 8 + FRAME_PIXELS)
if body is None:
break
frame_ix, ms = struct.unpack('<II', body[:8])
if last_ix is not None and frame_ix != last_ix + 1:
dropped += frame_ix - last_ix - 1
last_ix = frame_ix
f.write(struct.pack('<I', MAGIC))
f.write(body)
frames += 1
if frames % 25 == 0:
print(f'# {frames} frames, last ix={frame_ix} ms={ms} '
f'dropped={dropped}', file=sys.stderr)
print(f'# done: {frames} frames written to {args.out} '
f'({dropped} dropped)', file=sys.stderr)
if __name__ == '__main__':
main()

211
tools/replay_frames.py Normal file
View File

@@ -0,0 +1,211 @@
#!/usr/bin/env python3
# tools/replay_frames.py
#
# Offline Python port of the event-based CV detector (firmware/lib/cv/cv.cpp).
# Reads a .bin file produced by capture_frames.py and prints events.
#
# Purpose: iterate algorithm changes in seconds instead of minutes. All
# constants match cv.h so baseline behavior matches firmware.
#
# Usage:
# python tools/replay_frames.py walk.bin
# python tools/replay_frames.py walk.bin --enter 250 --exit 150 --max 25
#
# Output: one line per frame with fg diagnostics, plus [ENTRY]/[EXIT] lines
# when the detector fires.
import argparse
import struct
import sys
import numpy as np
MAGIC = 0x314D5246 # 'FRM1'
W = H = 96
PIXELS = W * H
HEADER = 12
FRAME_LEN = HEADER + PIXELS
class Detector:
"""Mirror of firmware CV state machine. Single walker events, centroid
trajectory direction. Only per-frame fg_count + min/max y + centroid y
feed the decision — per-blob tracks are diagnostic in firmware, dropped
here."""
def __init__(self, args):
self.a = args
self.bg = None
self.ev_active = False
self.ev_frames = 0
self.ev_first_c = -1.0
self.ev_last_c = -1.0
self.ev_min_c = float(H)
self.ev_max_c = -1.0
self.ev_min_y = H
self.ev_max_y = -1
self.ev_quiet = 0
self.last_fire = 0
self.frame_ix = 0
self.entries = 0
self.exits = 0
def _reset_event(self):
self.ev_active = False
self.ev_frames = 0
self.ev_first_c = self.ev_last_c = -1.0
self.ev_min_c = float(H)
self.ev_max_c = -1.0
self.ev_min_y = H
self.ev_max_y = -1
self.ev_quiet = 0
def _finalize(self):
a = self.a
if self.ev_frames < a.min_frames: return None
if self.ev_min_y > a.extent_top: return None
if self.ev_max_y < a.extent_bot: return None
up = self.ev_first_c - self.ev_min_c
down = self.ev_max_c - self.ev_first_c
winning = max(up, down)
if winning < a.min_traj: return None
is_entry = up >= down
self.last_fire = self.frame_ix
info = dict(
kind='ENTRY' if is_entry else 'EXIT',
first=self.ev_first_c, min=self.ev_min_c,
max=self.ev_max_c, last=self.ev_last_c,
dur=self.ev_frames,
)
if is_entry: self.entries += 1
else: self.exits += 1
return info
def step(self, frame):
"""frame: uint8 array of shape (H, W). Returns list of fire dicts."""
self.frame_ix += 1
fires = []
if self.bg is None:
self.bg = frame.astype(np.int16)
return fires
bg = self.bg.astype(np.int16)
diff = np.abs(frame.astype(np.int16) - bg)
fg = (diff > self.a.diff_thresh).astype(np.uint8)
# Running-avg bg blend, frozen during active event.
if not self.ev_active:
self.bg = ((self.bg * 31 + frame.astype(np.int16)) >> 5)
fg_count = int(fg.sum())
if fg_count > 0:
row_counts = fg.sum(axis=1)
ys = np.where(row_counts > 0)[0]
min_y = int(ys.min())
max_y = int(ys.max())
centroid_y = float((row_counts * np.arange(H)).sum() / fg_count)
else:
min_y, max_y, centroid_y = -1, -1, -1.0
# Self-heal on catastrophic bg mismatch.
if fg_count > PIXELS // 2:
self.bg = frame.astype(np.int16)
if self.ev_active: self._reset_event()
return fires
a = self.a
in_refractory = (self.last_fire != 0 and
(self.frame_ix - self.last_fire) < a.refractory)
if not self.ev_active:
if not in_refractory and fg_count >= a.enter_thresh:
self.ev_active = True
self.ev_frames = 1
self.ev_first_c = centroid_y
self.ev_last_c = centroid_y
self.ev_min_c = centroid_y
self.ev_max_c = centroid_y
self.ev_min_y = min_y
self.ev_max_y = max_y
self.ev_quiet = 0
else:
self.ev_frames += 1
if fg_count > 0:
self.ev_last_c = centroid_y
if centroid_y < self.ev_min_c: self.ev_min_c = centroid_y
if centroid_y > self.ev_max_c: self.ev_max_c = centroid_y
if min_y < self.ev_min_y: self.ev_min_y = min_y
if max_y > self.ev_max_y: self.ev_max_y = max_y
ended = False
if fg_count < a.exit_thresh:
self.ev_quiet += 1
if self.ev_quiet >= a.quiet_frames:
ended = True
else:
self.ev_quiet = 0
if self.ev_frames > a.max_frames:
ended = True
if ended:
fire = self._finalize()
if fire: fires.append(fire)
self._reset_event()
self.bg = frame.astype(np.int16)
return fires, fg_count, min_y, max_y, centroid_y
def iter_frames(path):
with open(path, 'rb') as f:
data = f.read()
n = len(data) // FRAME_LEN
for i in range(n):
off = i * FRAME_LEN
magic, ix, ms = struct.unpack('<III', data[off:off + HEADER])
if magic != MAGIC:
raise RuntimeError(f'bad magic at frame {i}: 0x{magic:08x}')
frame = np.frombuffer(data, dtype=np.uint8,
count=PIXELS, offset=off + HEADER).reshape(H, W)
yield ix, ms, frame
def main():
ap = argparse.ArgumentParser()
ap.add_argument('path')
ap.add_argument('--diff-thresh', dest='diff_thresh', type=int, default=30)
ap.add_argument('--enter', dest='enter_thresh', type=int, default=300)
ap.add_argument('--exit', dest='exit_thresh', type=int, default=200)
ap.add_argument('--quiet', dest='quiet_frames', type=int, default=3)
ap.add_argument('--min', dest='min_frames', type=int, default=5)
ap.add_argument('--max', dest='max_frames', type=int, default=25)
ap.add_argument('--extent-top', dest='extent_top', type=int, default=10)
ap.add_argument('--extent-bot', dest='extent_bot', type=int, default=85)
ap.add_argument('--min-traj', dest='min_traj', type=float, default=15.0)
ap.add_argument('--refractory', dest='refractory', type=int, default=15)
ap.add_argument('--quiet-log', action='store_true',
help='Suppress per-frame fg lines')
args = ap.parse_args()
det = Detector(args)
total = 0
for ix, ms, frame in iter_frames(args.path):
total += 1
out = det.step(frame)
if out == []:
if not args.quiet_log:
print(f'[{ix:4d}] bg init')
continue
fires, fg, miny, maxy, cy = out
if not args.quiet_log and fg > 0:
print(f'[{ix:4d}] n={fg:4d} y={miny:2d}..{maxy:2d} c={cy:5.1f}')
for fire in fires:
print(f' >>> {fire["kind"]} first={fire["first"]:.1f} '
f'min={fire["min"]:.1f} max={fire["max"]:.1f} '
f'last={fire["last"]:.1f} dur={fire["dur"]}')
print(f'\n# {total} frames entries={det.entries} exits={det.exits}')
if __name__ == '__main__':
main()

186
tools/replay_logs.py Normal file
View File

@@ -0,0 +1,186 @@
#!/usr/bin/env python3
# tools/replay_logs.py
#
# Replay the event state machine against text serial logs captured from the
# production firmware. Input lines of the form:
# [F] n=<fg_count> y=<min_y>..<max_y> c=<centroid_y>
#
# Those four values are exactly what the firmware's event state machine
# consumes — so we can iterate event-level params (thresholds, max_frames,
# extent gates, trajectory cutoffs, refractory) offline without needing raw
# frames or the device.
#
# Usage:
# python tools/replay_logs.py walk.log
# python tools/replay_logs.py walk.log --enter 250 --exit 100 --max 30 --min-traj 10
# cat walk.log | python tools/replay_logs.py - --ground-truth 12
import argparse
import re
import sys
LINE_RE = re.compile(
r'\[F\]\s+n=(?P<n>\d+)\s+y=(?P<miny>-?\d+)\.\.(?P<maxy>-?\d+)\s+c=(?P<c>-?\d+\.\d+)'
)
def parse_frames(text):
"""Yield (fg_count, min_y, max_y, centroid_y) per [F] line, in order."""
for line in text.splitlines():
m = LINE_RE.search(line)
if not m:
continue
yield int(m['n']), int(m['miny']), int(m['maxy']), float(m['c'])
class Detector:
"""Mirror of firmware event state machine. Only uses per-frame diagnostic
values — the same inputs the firmware feeds it."""
def __init__(self, a):
self.a = a
self.ev = False
self.ev_n = 0
self.ev_first = self.ev_last = -1.0
self.ev_min = 1e9
self.ev_max = -1.0
self.ev_miny = 1e9
self.ev_maxy = -1
self.ev_quiet = 0
self.last_fire = -10**9
self.ix = 0
self.entries = 0
self.exits = 0
self.fires = []
def _reset(self):
self.ev = False
self.ev_n = 0
self.ev_first = self.ev_last = -1.0
self.ev_min = 1e9; self.ev_max = -1.0
self.ev_miny = 1e9; self.ev_maxy = -1
self.ev_quiet = 0
def _finalize(self):
a = self.a
if self.ev_n < a.min_frames:
return ('reject_short', None)
if self.ev_miny > a.extent_top:
return ('reject_extent_top', None)
if self.ev_maxy < a.extent_bot:
return ('reject_extent_bot', None)
up = self.ev_first - self.ev_min
down = self.ev_max - self.ev_first
winning = max(up, down)
if winning < a.min_traj:
return ('reject_traj', None)
timed_out = self.ev_n > a.max_frames
if timed_out:
is_entry = self.ev_last < self.ev_first
else:
is_entry = up >= down
kind = 'ENTRY' if is_entry else 'EXIT'
self.last_fire = self.ix
info = dict(kind=kind, first=self.ev_first, min=self.ev_min,
max=self.ev_max, last=self.ev_last, dur=self.ev_n,
up=up, down=down, ix=self.ix)
if is_entry: self.entries += 1
else: self.exits += 1
self.fires.append(info)
return ('fire', info)
def step(self, n, miny, maxy, c):
self.ix += 1
a = self.a
refractory = (self.ix - self.last_fire) < a.refractory
if not self.ev:
if not refractory and n >= a.enter_thresh:
self.ev = True
self.ev_n = 1
self.ev_first = self.ev_last = c
self.ev_min = c; self.ev_max = c
self.ev_miny = miny; self.ev_maxy = maxy
self.ev_quiet = 0
return None
self.ev_n += 1
if n > 0:
self.ev_last = c
if c < self.ev_min: self.ev_min = c
if c > self.ev_max: self.ev_max = c
if miny < self.ev_miny: self.ev_miny = miny
if maxy > self.ev_maxy: self.ev_maxy = maxy
ended = False
if n < a.exit_thresh:
self.ev_quiet += 1
if self.ev_quiet >= a.quiet_frames:
ended = True
reason = 'quiet'
else:
self.ev_quiet = 0
if self.ev_n > a.max_frames:
ended = True
reason = 'timeout'
if ended:
result = self._finalize()
self._reset()
return (reason, result)
return None
def main():
ap = argparse.ArgumentParser()
ap.add_argument('path', help='log file, or - for stdin')
ap.add_argument('--enter', dest='enter_thresh', type=int, default=300)
ap.add_argument('--exit', dest='exit_thresh', type=int, default=200)
ap.add_argument('--quiet', dest='quiet_frames', type=int, default=3)
ap.add_argument('--min', dest='min_frames', type=int, default=5)
ap.add_argument('--max', dest='max_frames', type=int, default=25)
ap.add_argument('--extent-top', dest='extent_top', type=int, default=10)
ap.add_argument('--extent-bot', dest='extent_bot', type=int, default=85)
ap.add_argument('--min-traj', dest='min_traj', type=float, default=15.0)
ap.add_argument('--refractory', dest='refractory', type=int, default=15)
ap.add_argument('--ground-truth', type=int, default=0,
help='Total expected walks for accuracy calc')
ap.add_argument('-v', '--verbose', action='store_true',
help='Print every event end, including rejections')
args = ap.parse_args()
text = sys.stdin.read() if args.path == '-' else open(args.path).read()
det = Detector(args)
rejects = {}
for n, miny, maxy, c in parse_frames(text):
out = det.step(n, miny, maxy, c)
if out is None:
continue
reason, result = out
if result is None:
continue
kind, info = result
if kind == 'fire':
print(f' {info["kind"]:5} first={info["first"]:5.1f} '
f'min={info["min"]:5.1f} max={info["max"]:5.1f} '
f'last={info["last"]:5.1f} dur={info["dur"]:2d} '
f'exit={reason}')
else:
rejects[kind] = rejects.get(kind, 0) + 1
if args.verbose:
print(f' [drop {kind}]')
total = det.entries + det.exits
print(f'\n=== entries={det.entries} exits={det.exits} total={total} ===')
print(f'rejected events: {rejects}')
if args.ground_truth:
gt = args.ground_truth
acc = min(total, gt) / gt * 100
over = max(0, total - gt)
print(f'accuracy vs gt={gt}: {acc:.0f}% (over={over})')
if __name__ == '__main__':
main()