mad-dcoder/sentinel-trace

How mad-dcoder/sentinel-trace works

Sentinel-Trace is an investigator-driven CCTV loss-prevention tool for locating a person in a starting camera window, then following candidates across cameras through human-confirmed search rounds. Its workflow reads only requested video ranges, detects and tracks people, builds appearance embeddings and tracklets, ranks cross-camera matches, and stores case evidence and decisions. The dashboard supports review and report generation; documented capabilities include a learned camera-transition map and forensic DOCX export. Some links reflect README or SOURCE INDEX references because the corresponding implementation excerpts were not shown.

Main entry points: hunter.py On-demand hunt engine · settings.py Runtime settings · app.py Investigation dashboard · reid.py Appearance embeddings

Components 17

Investigation workflow

Investigation dashboard dashboard/app.py
On-demand hunt engine pipeline/hunter.py
Runtime settings config/settings.py

Video and vision

Video archive core/video.py
Motion sampling core/motion.py
Person detection core/detector.py
Tracklet construction core/tracklet.py
Clothing attributes core/attributes.py
Appearance embeddings core/reid.py
OSNet-AIN model core/models/osnet_ain.py

Cross-camera matching

Candidate matching core/matcher.py

Evidence and reporting

Case evidence store db/database.py
Learning and transitions db/database.py
Route inspection utils/inspector.py
Forensic report export utils/report_generator.py

Other components

Investigator
CCTV video exports

Connections 22

Investigator → Investigation dashboard: initiates review README.md
Investigation dashboard → On-demand hunt engine: starts and guides hunts dashboard/app.py
On-demand hunt engine → Video archive: reads requested windows pipeline/hunter.py
CCTV video exports → Video archive: provides recordings README.md
On-demand hunt engine → Motion sampling: checks for motion pipeline/hunter.py
On-demand hunt engine → Person detection: detects people pipeline/hunter.py
On-demand hunt engine → Tracklet construction: builds person tracks pipeline/hunter.py
On-demand hunt engine → Clothing attributes: ranks clothing clues pipeline/hunter.py
On-demand hunt engine → Appearance embeddings: extracts appearances pipeline/hunter.py
Appearance embeddings → OSNet-AIN model: runs model core/reid.py
On-demand hunt engine → Candidate matching: ranks candidates pipeline/hunter.py
On-demand hunt engine → Case evidence store: stores case evidence pipeline/hunter.py
On-demand hunt engine → Learning and transitions: learns decisions and transitions pipeline/hunter.py
Investigation dashboard → Route inspection: requests route review dashboard/app.py
Investigation dashboard → Forensic report export: generates report dashboard/app.py
Route inspection → Case evidence store: reads search evidence utils/inspector.py
Video archive → Runtime settings: loads camera configuration core/video.py
Person detection → Runtime settings: uses model settings core/detector.py
Motion sampling → Runtime settings: uses sampling settings core/motion.py
Tracklet construction → Runtime settings: uses selection settings core/tracklet.py
Appearance embeddings → Runtime settings: loads model settings core/reid.py
Candidate matching → Runtime settings: uses search settings core/matcher.py

Drawn by GitDiagram on November 13, 2026 from the repository's file tree, README and source files. Read as Markdown ·