Study On Application Of Fast Motion Estimation Algorithm Based On DSP

H.264 is characterized by low bit rate, high picture quality and high compression ratio, but it's high-efficient compression is at the cost of increase the complexity of coding algorithm. Throughout the whole coding process, motion estimation takes the longest time. There are many fast motion estimation algorithms used to solve this problem. But most of the existing fast motion estimation algorithms are for video images under normal circumstances, and often fail when used in specific applications. Problems of the existing motion estimation algorithm in the application of video surveillance are studied, and a solution to the problems is presented.Surveillance video usually has a relatively fixed background, changeless prospect content and relatively stable scene. It would waste lots of search time and computation time if use the general motion estimation algorithm and search in accordance with normal procedures and calculation. Combined with the characteristics of surveillance video, a fast motion estimation algorithm that fits the surveillance video is put forward according to the defect of present classic block-matching motion estimation algorithm in practical applications. This algorithm combines all-zero block predict algorithm and pre-termination algorithm, divide the video image into drastic motion block, mild motion block, slow motion block and all-zero block before motion search. Skip the all-zero block and use different search algorithm for the other three blocks. The algorithm can reduce motion estimation time while ensure the image quality, thus improve the encoding speed. Write code in VC platform and debug. Application feasibility of the algorithm is verified by comparing with the test data of UMHexagonS Algorithm.Transplant the algorithm to DM642EVM board, optimize the code according to the structural characteristics of DSP. By evaluation of processed image quality, computation cost and compression bit rate etc, when ensure the video quality, this algorithm is proved to have significant improvement in motion estimation and save the coding time compared with UMHexagonS algorithm. This algorithm is more suitable for the monitor video with stable background and changeless prospect and can be used in surveillance video encoder design.