基于候选区域检测的核相关目标跟踪算法

Kernel Correlation Target Tracking Algorithm Based on Candidate Area Detection

近年来,核相关滤波算法在目标跟踪领域应用广泛,表现出了非常优异的性能,但是核相关滤波类算法本质上属于模板匹配算法,并且缺乏跟踪失败恢复机制,在快速运动和快速形变情况下跟踪效果较差。针对以上问题,本文提出一种结合了核相关滤波跟踪算法和目标候选区域检测的跟踪算法,来改善核相关滤波跟踪算法的性能。算法主要设计了一种跟踪失败恢复机制,通过比较目标响应强度与经验阈值的大小,判断跟踪目标是否跟丢,当目标跟踪失败时,采用候选区域检测算法,在目标周围区域提取不同的检测图像块,确定目标在当前帧的最佳位置;然后,使用核相关滤波算法得到目标的精确位置,继续跟踪。此外,算法在跟踪模块中加入了颜色特征与梯度特征的自适应融合,进一步增强了算法的整体跟踪性能。实验结果证明,所提出算法在精确度和成功率上都表现出高效的性能,并且在快速运动和快速形变情况下跟踪性能要优于其余算法。

In recent years,the kernel correlation filtering algorithm has been widely used in the field of target traching and has shown very great performance.However,the kernel correlation filtering algorithm lacks the recovery mechanisrm of tracking failure.Under fast motion and rapid deformation,the result of tracking is not good.In view of the above problems,this paper presents a tracking algorithm combining kernel correlation filtering tracking algorithm and target candidate region detection to improve the performance of kernel correlation filtering tracking algorithm.The algorithm mainly designs a tracking failure recovery mechanism.By comparing the target response strength and the threshold of the experience threshold,the tracking target is judged whether or not the tracking target is lost.When the target tracking fails,the candidate area detection algorithm is adopted to extract different detection images in the area around the target block,to determine the target in the best position of the current frame;and then use the kernel correlation filtering algorithm to get the exact location of the target,and then continue tracking.In addition,the algorithm incorporates adaptive fusion of color features and gradient features into the tracking module,further enhancing the overall tracking performance of the algorithm.The experimental results show that the proposed algorithm shows high performance in accuracy and success rate,and the tracking performance is superior to the rest of the algorithms in fast motion and rapid deformation.