运动背景下的抗遮挡TLD改进算法

Improved anti-occlusion TLD algorithm under moving background

针对跟踪学习检测(Tracking Learning Detection,TLD)算法在目标受到遮挡、尺度变换和形变时跟踪精确度下降的问题,提出了运动背景下的抗遮挡TLD改进算法。在TLD算法内置的跟踪模块中引入面向加速分段测试特征和旋转二进制鲁棒独立基本特征(Oriented FAST and Rotated BRIEF,ORB)与均匀分布特征点相结合的特征点提取方法;在检测模块中采用自适应切换局部搜索与全局搜索的策略;采用警戒区域的方法进行遮挡判定,采用Kalman滤波与特征点匹配相结合的方法预测目标位置。实验结果表明,改进算法中跟踪模块中的特征点能够准确地反映目标的信息,同时减少了目标检测的扫描区域,提高了对运动背景下被遮挡目标的判定及预测能力。与同类目标跟踪算法相比,该算法跟踪速度、精度都较高;与经典TLD算法相比,改进算法在面对运动遮挡场景时表现更好。

For the problem that the tracking accuracy of the tracking learning detection(TLD)algorithm declines when the target is occluded,scaled or deformed,an improved anti-occlusion TLD algorithm under moving background is proposed.In the tracking module built in the TLD algorithm,a feature point extraction method that combines the oriented FAST and rotated BRIEF(ORB)and uniformly distributed feature points is introduced;In the detection module,the strategy of adaptive switching between local search and global search is adopted;The method of warning area is used to judge the occlusion,and the combination of Kalman filter and feature point matching is introduced to predict the target position.Experiment results show that the feature points in the tracking module of the algorithm can accurately reflect the information of the target,reduce the scanning area of the target detection,and improve the judgment and prediction ability of the blocked target in the moving background.Compared with similar target tracking algorithms,it has higher tracking speed and accuracy.Compared with the classical TLD algorithm,it performs better in occlusion scenes.