This is our implementation of a real-time CCTV shoplifting detection system.
We used the videos in the shoplifting category of the UCF-Crime Dataset.
Each video was split into 4 second clips using FFmpeg. The clips were then overlayed with pose information using the YOLOv11 model
Frames of the updated clip are then used as input to a 3D CNN which outputs the shoplifting probability of the entire clip