针对视频图像,提出了一种两步二进制位操作的目标识别跟踪算法。该算法基于改进的二进制鲁棒角点算法(BRISK),对特征描述符建立方法及匹配算法进行了改进:提出了采样点对选择策略和幅值-旋转两级描述符的建立方法;在特征点匹配阶段,提出“移位”结合“异或”的两步位操作特征匹配算法,并通过部分匹配及检测汉明重量阈值的方式进一步加快算法执行速度。实验结果表明,该改进方法具有较快的运算速度,目标平均跟踪速度达到80fps以上,且内存需求量小,更好的满足了视频图像目标实时识别跟踪的应用要求。

Aiming at the target recognition and tracking within video image, a combined two step bits operation algorithm which based on an improved binary robust invariant scalable keypoints (BRISK) algorithm was proposed. The improvements include descriptor building and feature matching methods. During descriptor building, a point pair selection strategy for reducing relativity and scope-rotation two steps feature descriptor were put forward. In feature matching stage, the algorithm which combined'SHIFT''XOR' binary operations was used. To further improve feature matching speed, an accelerated method with partial matching and Hamming weight threshold checking was introduced. The result of experiments demonstrate that the proposed algorithm achieves the average target tracking speed of 80 frames per-second, which is faster than other corner feature algorithms. Therefore, it could meet the demand of real-time target recognition and tracking from video image more effectively.