Documentation
# TraceAI — AI-Assisted CCTV Investigation Platform
TraceAI is an AI-assisted investigation platform designed to help authorized investigators trace a missing person across large amounts of disconnected CCTV footage.
## The Problem
When a person goes missing, investigators may need to manually review hours of footage from multiple cameras. Different camera locations, timestamps, and disconnected recordings make it difficult to reconstruct a person's possible movement.
TraceAI is designed to make this investigation process faster and more structured.
## How It Works
An investigator begins by uploading:
- A legally obtained reference photograph
- CCTV footage from authorized sources
- Camera and location information
The AI pipeline can then:
1. Detect people in CCTV footage
2. Track individuals across frames
3. Generate person appearance embeddings
4. Perform Person Re-ID across different camera footage
5. Search visually similar tracks using vector search
6. Generate possible sightings
7. Organize sightings chronologically
8. Visualize possible movement on a map
Each possible sighting contains:
- Camera
- Location
- Timestamp
- CCTV frame
- Track ID
- Similarity score
- Review status
## Human-in-the-Loop Investigation
TraceAI does not automatically identify or label someone as a suspect.
Instead, AI results are presented as:
Possible Match
Visual Similarity
Requires Human Verification
Investigators remain responsible for reviewing evidence and deciding whether a sighting should be added to the investigation timeline.
## Investigation Workflow
Reference Photo
→ CCTV Sources
→ AI Processing
→ Possible Sightings
→ Evidence Review
→ Timeline
→ Movement Map
## Version 2
The future version can support authorized live CCTV/RTSP sources.
When a possible visual match appears, investigators can receive an alert and review the associated camera frame.
## Design Goal
The interface combines:
- Ethiopian Federal Police-inspired institutional identity
- CCTV investigation
- AI-assisted search
- Evidence review
- Timeline reconstruction
- Geographic visualization
The goal is to create an interface that feels like a professional investigation workstation rather than a generic AI dashboard.
## Important Principle
Similarity is not identity.
The system is designed to assist investigators in searching and organizing evidence, while keeping human verification at the center of the investigation.
