Real-time moving object detection in H.264 encoding domain

A novel real-time moving object detection scheme in H.264 encoding domain is proposed. This scheme employs the encoding information of motion vectors (MVs) and the sum of absolute differences (SADs) as the input features to reduce the computational complexity, and takes the global motion into consideration for the video scenes with moving backgrounds. First, a fast global motion estimation model and a spatiotemporal filter for MVs are combined to detect the moving regions with salient motion. Second, χ2 distribution of the SAD at zero MV is constructed, and then a change detection based on the F hypothesis test is presented to detect the moving regions with salient motion and non-salient motion. Finally, the detected results of the above two steps are post-processed and utilized to compute the final moving object region map. Experimental results show that the proposed scheme is effective and efficient.