How poojitha20-b/illegal_dumping works
VidTrace turns live or recorded CCTV footage into confirmed dumping verdicts and enforcement actions. The documented workflow runs perception, identity-preserving tracking, temporal memory and feature extraction, event inference, then agentic review; confirmed cases proceed to plate OCR or pedestrian face identification, challan creation, and notification. A Flask dashboard exposes violations and hotspots. The README documents this wider flow, while sampled source confirms selected components and runner actions; arrows for unsampled internals follow documented behavior rather than inspected implementation.
Main entry points: run_pipeline.py Pipeline runner · app.py Review dashboard · hotspot_manager.py Hotspot manager · agent.py Belief agent
Components 23
Video perception
- Object detection
Layer1/detector.py - Trash detection
Layer1/trash_detector.py - Bin detection
Layer1/bin_detector.py - Object tracking
Layer2/tracker.py - Bin tracking
Layer2/bin_tracker.py
Temporal reasoning
- Temporal memory
Layer3/memory_manager.py - Behavior features
Layer3/feature_extractor.py - Dumping inference
Layer4/dumping_inference.py - Belief agent
Layer5/agent.py - LLM controller
Layer5/llm_controller.py
Enforcement
- Plate OCR
enhancer.py - Face identification
FaceID/face_id_module.py - Offender database
FaceID/database.py - Challan manager
penalty_manager.py - Email delivery
delivery_agent.py
Review and records
- Hotspot manager
hotspot/hotspot_manager.py - Hotspot store
hotspot/hotspot_manager.py - Review dashboard
dashboard/app.py
Other components
- CCTV / video
- Pipeline runner
run_pipeline.py - Groq service
- SMTP service
- Enforcement reviewer
Connections 23
- CCTV / video → Pipeline runner: provides footage
- Pipeline runner → Object detection: detects objects
- Pipeline runner → Trash detection: detects trash
- Pipeline runner → Bin detection: detects bins
- Pipeline runner → Object tracking: tracks objects
- Pipeline runner → Bin tracking: tracks bins
- Pipeline runner → Temporal memory: updates history
- Pipeline runner → Behavior features: extracts behavior
- Pipeline runner → Dumping inference: evaluates events
- Pipeline runner → Belief agent: reviews candidates
- Belief agent → LLM controller: requests verdict
- LLM controller → Groq service: calls model
- Pipeline runner → Plate OCR: resolves plates
- Pipeline runner → Face identification: identifies pedestrians
- Face identification → Offender database: matches / records
- Pipeline runner → Challan manager: creates challan
- Pipeline runner → Email delivery: sends notice
- Email delivery → SMTP service: sends email
- Pipeline runner → Hotspot manager: records location
- Hotspot manager → Hotspot store: reads / writes
- Enforcement reviewer → Review dashboard: reviews cases
- Review dashboard → Challan manager: reads violations
- Review dashboard → Hotspot manager: reads hotspots