基于单类支持向量机的目标跟踪算法研究

Object Tracking Algorithm Based on One-class SVM Method

基于分类的跟踪算法成为当前目标跟踪的研究热点。首先把跟踪问题看成是一个目标和背景的二分类问题,根据每一帧的正负样本数据训练SVM分类器,通过分类器的分类概率值确定目标位置。然而,采集正负样本边界的那些样本很容易出现异常点,当把它们作为目标的下一帧位置时将会出现严重的跟踪漂移问题。为此,提出了一种基于单类支持向量机(One-class Support Vector Machine,One-class SVM)的目标跟踪算法,基于One-class SVM分类能有效地排除其他类的干扰,有效地防止异常样本的出现。并结合加权多示例采样方法,使得每个采样样本会根据不同的权值对于分类器的贡献而不同。仿真实验结果表明,改进的跟踪方法是可行的、有效的,有很好的鲁棒性。

Object tracking algorithm based on binary classification has become the research hot issues. The tracker is firstly as binary classification between object and background, and both positive and negative samples data are applied to train SVM classifier. The object＇ s location is determined by the probability of the classifier. However, such binary classification may not fully handle the oufliers, which may cause tracking drifting. To im- prove the robustness of the tracker, a novel object tracking algorithm is proposed based on one-class SVM. This method based on one-class support vector machine （SVM） can effectively rule out other types of interference, to effectively prevent the emergence of abnormal samples. Furthermore, the tracker integrates weighted multi-instance sampling method, which can consider the sample importance by the different weights. The experimental re- sults show the robustness of the improved method.