基于高效卷积算子的异常抑制目标跟踪算法

Anomaly Suppression Target Tracking Algorithm Based on Efficient Convolution Operator

传统的目标跟踪算法易受边界效应影响,且当目标因遮挡严重、运动模糊、光照变化等产生外观变化时,目标响应图会发生突变,从而降低目标跟踪检测结果的可信度。提出一种改进的高效卷积算子(ECO)目标跟踪算法。利用高斯混合模型生成紧凑且多样化的样本数据,采用因式分解卷积方法减少模型参数,引入空间权值系数和前后两帧响应图的变化率来弱化边界效应并抑制响应图突变,以提高目标跟踪算法的鲁棒性能和精度。实验结果表明,在光照、尺度变化等多种干扰下,该算法的成功率和距离精度较原始ECO算法分别提高3.1个百分点和1.9个百分点。

The traditional target tracking algorithm is easily affected by the boundary effect,and its target response graph mutates with the target appearance changes caused by occlusion,motion blur and varying illumination,which reduces the credibility of the target tracking and detection results.An improved target tracking algorithm based on Efficient Convolution Operator(ECO)is proposed.The algorithm employs the Gaussian mixture model to generate compact and diverse sample data,and uses the factorization convolution method to reduce the number of parameters.Then the spatial weight coefficient and the change rate of the two response graphs are introduced to diminish the boundary effect and suppress the mutation of the response graph,so as to improve the robustness and accuracy of the target tracking algorithm.Experimental results show that the success rate and distance accuracy of the proposed algorithm are 3.1 percentage points and 1.9 percentage points higher than those of the original ECO algorithm respectively under various interference such as illumination and scale change.