Project Case Study
AI Security Investigator
A CCTV investigation system for uploading footage, reviewing detected and tracked objects, searching for events, and assembling case evidence.
Overview
AI Security Investigator is a CCTV investigation system that connects video analysis to case review, search, audit, and evidence handling.
Problem
Investigators need to find relevant events across video and camera timelines, then preserve context while reviewing and sharing evidence.
Architecture
- 01
Video upload
- 02
Frame processing / YOLOv8 detection
- 03
Tracking + event extraction
- 04
FastAPI / SQLAlchemy persistence
- 05
Search, cases, audit, evidence
The flow follows the repository's video processor, detection/tracking modules, database, forensic API, and investigation UI. Optional AI/vector services are not required for the main workflow.
What I Built
- Video upload and processing with YOLOv8 detection and selectable lightweight or DeepSORT tracking.
- Investigation search, video/case workflows, multi-camera timelines, and audit records.
- Evidence clip export with a JSON sidecar and SHA-256, plus redaction and case-report workflows.
- Optional Gemini/Ollama summaries and optional ChromaDB vector search.
Technical Challenges
Connecting frame-level detections and tracks to timestamped events, multi-camera case timelines, search results, and reviewable evidence.
Stack
Validation / Testing
- Repository includes backend API tests and documents a forensic smoke-test script.
- No detection-accuracy or incident-resolution benchmark is claimed.
Current Status
Prototype. The documented Hugging Face Space is paused, and a working public deployment was not verified.