Real-time CCTV vehicle monitoring

Problem

Live CCTV vehicle monitoring needed detection that could keep up with streaming.

Approach

RTSP ingest through ffmpeg, 2 YOLO models in parallel processes, decoupled with a Redis queue. Records are stored in S3.

Outcome

Working prototype. It ingests a real CCTV camera streaming over RTSP, runs 2 YOLO models in parallel processes, uses a Redis queue, stores records in S3, and deployed with Docker.

Amber flowchart on a black field: live camera, capture and detect vehicles, no vehicle exits while vehicle detected goes to drop batch, then Redis queue and temp store, track and capture stills, and S3 image store.

Skills used

Python multiprocessing
YOLO
Redis
ffmpeg
S3