Project Overview
Detectorcam: Real-Time Object Detection & Tracking System
Detectorcam is a web-based system for real-time object detection and tracking from RTSP and HTTP video streams. The system integrates the YOLO detection model, BoT-SORT/ByteTrack tracking algorithm, and stream health monitoring mechanisms to ensure smooth video processing in environments with unstable connections.
Core Features
Object Detection & Tracking
- Utilizes the YOLO model (
.pt format) for object detection.
- Supports switching models for flexibility in change other model, accuracy, and performance.
- Implements BoT-SORT/ByteTrack from Ultralytics to assign unique IDs to detected objects across frames.
Video Streaming & Stability Handling
- Monitoring RTSP connection and automatically reconnecting or restarting if disconnect, lag, buffering, or stream freeze is detected.
- Implementing frame skipping to maintain target FPS and responsiveness by skipping frames when inference time exceeds the limit.
Web Interface
- CCTV Management: Add, remove, and configure CCTV streams.
- Detector & Model Management: Manage detection models and assign them to streams.
- Live Streaming:
- Camera Stream: Displays the original video.
- Detector Stream: Displays annotated video with detection and tracking results.
- Mode Control: Enable or disable object detection/tracking for each stream directly from the interface.
Key Achievements
- ID Tracking: Integrated BoT-SORT/ByteTrack for robust object ID tracking across frames.
- Connection Stability: Implemented auto-reconnect and lag detection for unstable RTSP connections.
- Performance Optimization: Developed frame skipping mechanism to maintain target FPS under heavy load.
- Control Dashboard: Delivered a web-based interface with real-time detection, tracking, and stream health monitoring controls.
