Suriyakumarvijayanayagam
dad04e8cda
Five components that ship as one product:
- behavision/ the recognition engine. RTSP ingest, YuNet detection, IoU
tracking, ArcFace embeddings, a FAISS/SQLite gallery, and a
FastAPI dashboard. Identity is decided once per TRACK from an
average of at least three embeddings, never per frame.
- agent/ the Go edge agent: supervises the engine, holds a durable
spool, and drains it to MQTT. Nothing is acked before the
broker confirms.
- desktop/ the shop PC application (Wails + React + tray).
- server/ the cloud API, MQTT consumer, reports and assistant.
- web/ platform.loyaly.ai, the head-office app, embedded in the
server binary.
The gallery stores 512-float embeddings and timestamps - no images unless
`app.store_faces` is switched on. Those embeddings are biometric personal
data under GDPR and India's DPDP: template inversion reconstructs a
recognisable face from an ArcFace vector, so data/behavision.db is treated
as a biometric database and DELETE /api/visitors/{id} is a real erasure.
CLAUDE.md carries the reasoning behind every non-obvious decision here,
including the ones that were measured and the ones that were wrong first.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01HViLj9gYNRtSr7YVZmW5sn
92 lines
3.5 KiB
Python
92 lines
3.5 KiB
Python
"""Attribute aggregation — pure maths, no models."""
|
|
from behavision.attributes import aggregate
|
|
|
|
|
|
def test_median_age_ignores_one_wild_frame():
|
|
samples = [{"age": 47}, {"age": 51}, {"age": 12}] # 12 = a bad frame
|
|
out = aggregate(samples)
|
|
assert out["age"] == 47 # median, not mean (mean would be 36)
|
|
assert out["age_spread"] == 39 # and the disagreement is reported
|
|
|
|
|
|
def test_majority_gender_wins_over_a_single_flip():
|
|
out = aggregate([
|
|
{"gender": "Male", "gender_confidence": 0.9},
|
|
{"gender": "Male", "gender_confidence": 0.8},
|
|
{"gender": "Female", "gender_confidence": 0.99},
|
|
])
|
|
assert out["gender"] == "Male"
|
|
assert out["gender_confidence"] == 0.85 # mean over the winning frames only
|
|
|
|
|
|
def test_empty_and_partial_inputs():
|
|
assert aggregate([]) == {}
|
|
assert aggregate([{}, None]) == {}
|
|
assert aggregate([{"gender": "Female", "gender_confidence": 0.7}]) == {
|
|
"gender": "Female", "gender_confidence": 0.7}
|
|
|
|
|
|
def test_caffe_age_range_is_aggregated_too():
|
|
out = aggregate([{"age_range": "38-43"}, {"age_range": "38-43"},
|
|
{"age_range": "25-32"}])
|
|
assert out["age_range"] == "38-43"
|
|
|
|
|
|
def test_person_event_reports_the_quality_the_gate_actually_used():
|
|
"""The enrollment gate runs on track.best_quality. Reporting this frame's
|
|
track.quality instead made events look like they had passed a threshold
|
|
they were below (observed: quality=0.586 logged against a 0.65 gate)."""
|
|
import threading
|
|
|
|
import numpy as np
|
|
|
|
from behavision.config import Config
|
|
from behavision.engine import CameraWorker
|
|
from behavision.gallery.service import Resolution
|
|
from behavision.faces import FaceOutbox
|
|
from behavision.tracking import Track
|
|
|
|
cfg = Config()
|
|
published = []
|
|
|
|
class Enc:
|
|
size = 112
|
|
def encode_chip(self, chip):
|
|
v = np.ones(512, np.float32)
|
|
return v / np.linalg.norm(v)
|
|
|
|
class Gal:
|
|
def resolve(self, mean, quality, cam, ts, attributes=None, rcfg=None):
|
|
self.seen_quality = quality
|
|
self.seen_rcfg = rcfg
|
|
return Resolution(kind="new", identity_id=1, label="Visitor 1",
|
|
similarity=0.1, new_sighting=True)
|
|
|
|
w = object.__new__(CameraWorker)
|
|
w.cfg, w.encoder, w.attrs = cfg, Enc(), None
|
|
w.rcfg = cfg.recognition.merged(None) # no per-camera overrides here
|
|
w.commission = None # no placement check running
|
|
w.gallery = Gal()
|
|
w.cam_cfg = type("C", (), {"id": "cam1"})()
|
|
w.bus = type("B", (), {"publish": staticmethod(published.append)})()
|
|
# Images are off, the product default. A real FaceOutbox rather than a
|
|
# mock, so a worker built this way runs the same disabled path
|
|
# production does when store_faces is unset.
|
|
w.faces = FaceOutbox(cfg.app.data_dir, enabled=False)
|
|
|
|
kps = np.array([[130, 100], [190, 100], [160, 130], [135, 165], [185, 165]],
|
|
np.float32)
|
|
track = Track(id=1, box=(100, 60, 220, 200), kps=kps, score=0.9,
|
|
quality=0.40, best_quality=0.80, hits=10)
|
|
frame = np.zeros((240, 320, 3), np.uint8)
|
|
ts = 1000.0
|
|
for _ in range(cfg.tracking.min_embeddings_for_id):
|
|
ts += 0.04
|
|
w._identify(track, frame, ts)
|
|
|
|
assert published, "no event published"
|
|
data = published[0].data
|
|
assert w.gallery.seen_quality == 0.80 # the gate saw best_quality
|
|
assert data["quality"] == 0.80 # and so does the event
|
|
assert data["frame_quality"] == 0.40 # this frame, kept for context
|