特征融合与重定位卷积算子跟踪算法研究

针对卷积操作目标跟踪算法(ECO-HC)在遮挡、背景等干扰问题导致跟踪精度下降的问题,提出了一种自适应特征融合的卷积相关滤波算法,将CN与HOG特征进行加权融合,通过计算各自的响应来确定各自特征在下一帧的权重,将特征各自的优势充分发挥出来.此外,针对目标跟踪失败问题,提出利用形变相似多样性原理,构建目标重定位模块,当出现遮挡、快速移动等复杂情况造成跟踪的可靠性降低时,综合考虑目标响应得分、空间权重得分和形变相似多样性得分来确定目标的最终位置,实现重定位.实验证明,改进后算法与ECO-HC相比,针对目标遮挡、背景干扰等复杂情况,有效地提高了跟踪精度,鲁棒性更强.

Research on tracking algorithm of feature fusion and relocation convolution operator

Aiming at the problem of efficient convolution operators based tracking algorithm fusing the histogram of oriented gradient and color names features(ECO-HC) in the occlusion, background and other interference problems leading to the reduction of tracking accuracy, this paper proposed an adaptive feature fusion convolution correlation filter algorithm, which combined the CN feature and the HOG feature by weighted fusion. Determine the weight of each feature in the next frame by calculating the respective response, and gave full play to the advantages of each feature. In addition, for the problem of target tracking failure, this paper proposed using the principle of deformation similarity diversity to construct a target relocation module. When complex conditions such as occlusion and rapid movement cause the reliability of tracking to be reduced, considering the target response score, spatial weight score and deformation are comprehensively to determine the final position of the target and achieve relocation. Experiments show that compared with ECO-HC, the improved tracking accuracy is more effective for target occlusion, background interference and other complex situations, and it is more robust.