Author: Divyanshi Singh
Project Type: Academic Research & Development
Traditional CCTV systems are entirely passive, relying on post-incident review and the limited attention span of human operators monitoring multiple screens. This project transforms video surveillance from a passive recording tool into an active, real-time threat detection system.
This repository contains an end-to-end computer vision pipeline designed to automatically detect and track theft or shoplifting behaviors in live CCTV feeds. By combining state-of-the-art object detection with advanced tracking algorithms, the system can identify suspicious interactions between individuals and high-value items, maintaining track of suspects even through occlusions.
- Real-Time Processing: Optimized for live CCTV feeds to maintain high Frames Per Second (FPS).
- High-Accuracy Detection: Utilizes YOLOv8 for single-pass inference to identify persons, bags, hands, and merchandise.
- Persistent Object Tracking: Integrates DeepSORT to assign and maintain unique IDs for individuals, preventing ID switching when suspects cross paths or walk behind shelves.
- Behavioral Logic Detection: Custom heuristics to flag suspicious temporal sequences (e.g., a hand interacting with an item, followed by the item disappearing into a bounding box for a bag/pocket).
- Video Input: Live stream or recorded footage processed via OpenCV.
- Detection: YOLOv8 extracts precise bounding boxes for target classes.
- Tracking: DeepSORT applies Kalman filtering and deep appearance features for ID persistence.
- Action Recognition: The system analyzes spatial intersections over a set frame window.
- Alerting: An automated flag is triggered upon a positive theft classification.
- Python 3.8+
- Git
- Clone the repository:
git clone [https://github.com/yourusername/ai-theft-detection-cctv.git](https://github.com/yourusername/ai-theft-detection-cctv.git) cd ai-theft-detection-cctv