Harris角点结合金字塔光流法的目标跟踪算法设计研究

Design and research of objective tracking algorithm using Harris corner and Pyramid optical flow method

摘要:

针对现存很多跟踪算法在速度和准确度方面很难满足嵌入式跟踪开发的需要,提出一种基于Harris角点和金字塔光流法的快速跟踪算法,并详细给出了DSP-FPGA的硬件设计。首先,使用Harris角点提取目标角点特征；然后,使用金字塔光流法为后续视频帧匹配角点；最后,基于角点的质心跟踪算法用于匹配目标的重心,确定目标的位置,重心跟踪算法可以较好地抵消由于旋转或扭曲带来的形变问题。在硬件实现过程中,FPGA方便电路设计,使用硬件描述程序语言实现硬件算法、逻辑控制和外部接口,DSP则运行目标跟踪算法。实验结果验证了本文硬件实现算法的有效性,相比于AVT21开发板的质心跟踪算法、相位相关跟踪算法和金字塔相关性跟踪算法相比,本文算法在平均重叠和平均中心误差方面具有一定优势,在720p的视频流上可以满足25fps。

Abstract:

For many existing tracking algorithm is difficult to meet the development needs of embedded tracking in terms of speed and accuracy, fast tracking algorithm is proposed based on Harris corner and Pyramid optical flow, with details on DSP-FPGA hardware design given. Firstly, Harris corner is used to extract feature points of targets. Then, Pyramid optical flow is used to match corner for subsequent video frame. Finally, centroid tracking algorithm based on corner is used to match the center of gravity, determining the location of the target, better counteracting distortions caused by the rotation or deformation. In the hardware implementation process, FPGA convenient circuit design using a hardware description language program hardware algorithms, logic control and external interface, and DSP runs target tracking algorithm. The effectiveness of the proposed algorithm implemented in hardware is verified by experimental results. Compared to trackers in AVT21 development board, centroid algorithm, phase correlation tracking algorithm and pyramid correlation tracking algorithm, the proposed algorithm has more advantages in means of overlapping and means of center error. In 720p video stream, the video can satisfy 25fps.