为降低序贯相似性检测算法(SSDA)的计算量,提出了一种双阈值快速SSDA算法(DTSSDA)。DTSSDA增加了误差累积速度阈值,当误差累积速度超过速度阈值时可提前终止本次匹配计算,而速度阈值可利用帧间相关性信息自动地选取。另外算法中还引入了积分图的概念并采用了从粗到精的搜索策略。实验表明DTSSDA能在保证匹配精度基本不变的前提下使单帧计算量大幅减少,因此更适合于实时目标跟踪。

In order to reduce the computational cost of sequence similarity detection algorithm (SSDA), a fast SSDA with double thresholds (DTSSDA) was proposed in this paper. DTSSDA introduces another threshold related with error-increasing-speed (EIS), when EIS of a reference region exceeds the threshold, operations cease for the region. Threshold of EIS can be determined adaptively according to correlation of inter-frame. Integral image was introduced in and search strategy of rough to fine is adopted in the algorithm. Experimental results show DTSSDA can reduce the computational cost in large scale and maintain almost the same accuracy as SSDA, so it is more suitable for real-time target tracking.