融合重检测机制的上下文感知目标跟踪算法

Context-aware target tracking algorithm fused with redetection mechanism

为了解决目标跟踪中常见的遮挡、旋转和背景杂乱等问题,提出了一种融合重检测机制的上下文感知目标跟踪算法。首先在相关滤波算法的基础上引入上下文信息供滤波器学习以丰富样本信息,构造上下文感知相关滤波器,提高滤波器的学习能力;然后引入重检测机制判断检测结果的可靠性,解决遮挡情况下模型被污染的问题。最后在公开数据集上对算法的性能进行了测试,并与DSST、Staple、SRDCF、TLD和BACF这5种算法进行对比。实验结果表明,算法在遮挡、旋转和背景杂乱等复杂场景下具有较好的跟踪鲁棒性,跟踪精度和成功率分别达到了0.748和0.836,均优于其余5种跟踪算法。

In order to solve the common problems of occlusion,rotation and background clutter in target tracking,a context-aware target tracking algorithm with re-detection mechanism is proposed in this paper.To solve the problem,first of all,the context information is introduced on the basis of the correlation fitering algorithm,with the purpose of making the filter to enrich the sample information.Furthermore,the context-aware correlation filter is constructed to improve the learning ability of the filter.And then,a re-detection mechanism is introduced to judge the reliability of the detection result,so that the problem of model contamination in the case of occlusion can be solved.Finally,,the performance of the proposed algorithm is tested on the public datasets and compared with the five algorithms,including DSST(Accurate Scale Estimation for Robust Visual Tracking),Staple(Complementary Learners for Real-Time Tracking),SRDCF(Learning Spatially Regularized Correlation Filter for Visual Tracking),TLD(Tracking-Learning-Detection)and BACF(Learning Background-Aware Correlation Filter for Visual Tracking).The experimental results show that the algorithm in this paper has better tracking robustness in complex scenes such as occlusion,rotation and background clutter.The tracking accuracy and success rate of the algorithm have reached 0.748 and 0.836 respectively,which are better than the other five tracking algorithms.