引入特征重检的抗遮挡目标跟踪方法研究

Research on Anti-occlusion Target Tracking Method with Feature Redetection

针对视觉目标跟踪的遮挡问题,在TLD算法的基础上,引入特征重检环节,解决发生遮挡时因目标外观相似、背景聚类造成错判,提出了一种基于特征重检的抗遮挡目标跟踪研究方法(TLD-D),采用跟踪、检测、学习、再检测的策略。跟踪与检测相结合,对锁定的目标进行学习,获取目标最新的外观特征;当发生遮挡时,则启用特征重检环节,提取遮挡过程的"开始发生遮挡"和"遮挡结束"两个关键帧,然后在特征重检环节选用SIFT特征进行双向匹配标定目标,确保重新标定的目标为原被遮挡的跟踪目标,即"再检测"。OTB基准集上实验结果表明,与TLD算法、同类TLD改进算法以及其他经典跟踪算法相比较,TLD-D算法抗遮挡能力更强,鲁棒性更强,能够对目标长时间稳定跟踪。

Aiming at the occlusion problem of visual target tracking,this paper introduces feature redetection based on TLD(Tracking-Learning-Detection)algorithm to solve the misjudgments caused by similar appearance and background clustering of the target when occlusion occurs.A new method of anti-occlusion target tracking based on feature redetection(TLD-D)is proposed,which adopts the strategies of tracking,detection,learning and redetection.When occlusion occurrs,feature redetection is enabled to extract the two key frames of occlusion process,which are called the beginning of occlusion and the end of occlusion,and then SIFT feature is selected in feature redetection for bidirectional matching to calibrate the target to ensure that the re-calibrated target is the original tracking target,namely"redetection".The experimental results show that compared with TLD algorithm and other classical tracking algorithms,the TLD-D algorithm has stronger anti-occlusion ability and robustness,and can track the target stably for a long time.