A video codec based on background extraction and moving object detection

Cameras are the primary data sources in video surveillance systems and produce massive data every second. Video surveillance is an extremely beneficial functionality brought to us by modern technology. An essential application of video surveillance in public security is facilitating the observation and analysis of events. Video surveillance systems require high-bandwidth media to transfer, high-capacity media to store, and high-performance hardware to process data. Consequently, these systems impose many costs on organizations. Video compression techniques can reduce the amount of data transferred or stored by surveillance systems and, as a result, lower the costs. Fixed CCTV cameras are the largest category of surveillance cameras. Backgrounds in these videos are typically constant and saving them for every frame is redundant. Therefore, a background-aware approach can achieve a higher compression rate in compressing these cameras' videos than conventional approaches. This paper proposes a video codec for fixed cameras based on background extraction and moving-object detection algorithms. By background extraction, the pure backgrounds of the video are generated and stored in JPEG format for consecutive time intervals. By moving-object detection, the objects, and their coordinates are extracted in each frame using YOLOv7 and stored in JPEG format separately from the backgrounds. At the decoder side, each frame is built up using the generated background and detected objects which have been stored and transmitted as JPEG files. The evaluation of the proposed method on an appropriate set of videos from CDnet2014 and EWAP datasets shows that the proposed method can compress the videos effectively by significant compression ratios up to 46.76x, while the worst quality loss resulting from the compression is 0.99 according to the SSIM measure.