Efficient Processing of Moving Target Detection Based on SW26010

SW26010 is a nationally-produced processor developed by the National Parallel Computer Engineering Research Center. It has high concurrency. The Sunway TaihuLight System built by SW26010 has reached the top of the super calculation ranking many times. The Gaussian mixture model (GMM) is an effective method for extracting moving objects, but the traditional GMM has high hardware requirements and cannot meet real-time requirements. In this paper, we propose a new moving object detection acceleration method combined with SW26010 processor. First, we transplante the video codec library FFMPEG to SW26010 to support video stream extraction. Next, a sliding window mechanism is proposed on the single-core group to reasonably divide the image data blocks. And we create a double buffer between the master and slave cores to hide the communication delay. Further we improve the computing performance in parallel between multi-core groups. Finally, the experimental verification shows that our method can achieve satisfactory performance acceleration in practical applications. This design scheme has good scalability on the multi-processor group of the Sunway TaihuLight System.