Moving Object Detection in HEVC Video

Video surveillance has drawn much interest in monitoring physical assets, spaces and events over time for detection of threats as well as business and process monitoring purposes. However, the rising number of recorded videos has significantly increased the time and effort in manual event analysis and video content management. Therefore, automatic moving object detection is of great importance. Nowadays, for storage and transmission purposes, video usually appears in the compressed form. Therefore, in this paper, an automatic moving object detection method is proposed for HEVC video. Specifically, the number of bits spent on coding a frame, which can be extracted during encoding or retrieved from an encoded video bit stream, is exploited as the key feature for moving object detection. In addition, temporal sub-layering feature of HEVC is utilized to reduce the number of frame to be processed, which in turn magnifies the energy of the coded video frames without losing most of the predicted information. A coarse background / foreground mask is then formed based on bit consumption, and it is further refined via post processing to remove noise and to smooth the mask image. The proposed method achieves encouraging results in detecting slow moving objects, even with dynamic background.