stopsign_ai

Tracking stop sign behavior with a camera and AI

🌐 Live Demo: crestwoodstopsign.com

Objective:

• Monitor a street intersection with a camera streaming video via RTSP.
• Analyze the feed in real-time using AI to detect vehicles and evaluate their stop sign behavior.
• Store processed images and metadata for each vehicle pass.
• Display recent vehicle passes and statistics on a web interface.

Screenshot

System Components and Flow

1. RTSP Stream Capture (rtsp_to_redis.py)

• Function: Captures video frames from the RTSP IP camera.
• Process:
  • Connects to the RTSP stream.
  • Encodes frames as JPEG.
  • Pushes frames to Redis.

2. Video Analysis (video_analyzer.py)

• Function: Processes frames from Redis, performs object detection, tracking, and stop sign behavior analysis.
• Process:
  • Retrieves frames from Redis.
  • Uses YOLO AI model for vehicle detection.
  • Tracks detected vehicles and analyzes their stop sign behavior.
  • Annotates frames and pushes to new Redis queue.
  • Store data in Postgres, and images in self-hosted S3-style file store (MinIO).

3. FFmpeg Streaming (ffmpeg_service.py)

• Function: Converts processed frames into an HLS video stream for web display.
• Process:
  • Consumes annotated frames from Redis.
  • Feeds frames into FFmpeg to generate HLS stream.

4. Web Server (web_server.py)

• Function: Serves the web interface and live video stream.
• Features:
  • Displays live HLS video stream.
  • Shows recent vehicle passes with images and scores.
  • Integrates with Grafana for detailed statistics.

5. Monitoring and Metrics

• Prometheus Integration:
  • All services expose metrics to Prometheus for monitoring.
  • Metrics include frame processing times, FPS, memory usage, and more.
• Grafana Dashboards:
  • Visualize collected metrics for real-time monitoring and long-term trends.