一种基于标记修正和样本更新的目标跟踪算法

Object tracking based on label correction and sample updating

针对半监督自训练框架下进行目标跟踪的误差累积问题,提出一种结合标记修正与区域置信度样本更新的自适应跟踪算法。该算法将视频序列中的目标跟踪视为两类模式即目标与背景的分类问题,在半监督自训练框架下,选择SVM分类器分类目标与背景,结合K近邻和最小距离分类进行标记修正,并基于区域置信度提取新的样本更新分类器。实验结果显示,该方法有效改善了由于误差累积导致的漂移问题和目标遮挡后的跟踪失败。

To solve the error accumulation problem of object tracking in the framework of semi-supervised self-training,this paper proposed an adaptive tracking algorithm with label correction and sample updating.It treated the object tracking as a binary classification problem between objects and backgrounds,used SVM to classify the image block,corrected the classification result with K-nearest neighbor and minimum distance classification,and extracted new samples based on regional confidence to update the classifier.Experimental results show that the proposed method can effectively avoid the drift problem due to error accumulation and tracking failure since object occlusion.