This project Summarize hours of footage captured by static CCTV cameras into a concise clip that presents all events as if they are happening simultaneously, complete with timestamps. It utilizes video processing techniques, including YOLO-based object detection, custom tracking algorithms, and video summarization methods, to identify, track, and summarize moving objects efficiently
In today's world, many CCTV cameras are increasingly being used in homes for security. However, reviewing many hours of footage to spot suspicious activity can be a time-consuming and tedious process. Storing these lengthy videos also requires significant memory, often leading to reduced video resolution to save space. There is a need for a solution that can condense long footage into short clips while preserving key events.
- Video Preprocessing
- The video is read using OpenCV, and the frames per second (fps) are extracted. A background model is created using cv2.accumulateWeighted for overlaying tracked objects. Frames are processed one by one until the video ends.
- Object Detection
- The YOLOv8 model detects objects in each frame, filtering those of interest (e.g., people, cars) with a confidence score > 0.5. Bounding box coordinates and object labels are extracted for further processing.
- Object Tracking
- Objects are tracked across frames using spatial proximity and temporal continuity. A custom algorithm links bounding boxes, tracks movements, and handles object disappearance. Classes like Box and MovingObject store tracking data.
- Video Summarization
- A summarized video is created by overlaying tracked objects with labels and timestamps on the background frame. Frames are compiled into a short video using OpenCV's cv2.VideoWriter.
- Logging and Reporting
- A log tracks each object's type, appearance time, disappearance time, and duration. Logs are displayed in a UI and stored for analysis, with error handling for failed detections or frame reads.
