基于跟踪异常与相关性检验的目标丢失判断

Target Loss Judgment Method Based on Tracking Anomaly and Correlation Test

针对互补性实时跟踪算法(Staple)在目标丢失后不能察觉,提出了基于跟踪异常与相关性检验的目标丢失判断方法。在平均峰值相关能量的基础上通过对颜色直方图模型响应进行评估,提出了一种改进的跟踪置信度评估方法。根据跟踪置信度对跟踪状态进行评估,并在高置信度情况下使用目标区域构建目标相关性检验模板。当相关滤波模型响应置信度由低变高后,使用目标相关性检验模板与当前目标区域进行相关性检验得到相似度,根据相似度值大小判断目标是否丢失。在OTB-100标准数据集中选取22段视频进行验证,实验结果表明,所提出的方法在Staple算法跟踪过程中能够及时地检测出遮挡、出视野和光照变化等干扰因素导致的跟踪异常。能够正确地判断目标丢失,成功率达100%,为跟踪异常后是否进行目标重检测和实际工程应用中目标丢失判断提供可靠的依据。

In order to solve the problem that the Staple tracker is imperceptible after the target is lost,a target loss judg-ment method based on tracking anomaly and correlation test is proposed.By evaluating the response of the color histo-gram model based on the average peak to correlation energy,an improved tracking confidence evaluation method is pro-posed.Then,the tracking status is evaluated according to the tracking confidence.In the case of high confidence,the tar-get correlation test template is constructed by using the target area.When the confidence level of the correlation filter response changes from low to high,the similarity is obtained by performing a correlation test between the target correla-tion test template and the current target area.Target loss is determined by similarity.The experiment is conducted on 22 video sequences of OTB-100 standard dataset.Experimental results show that the tracking anomalies caused by interfer-ence factors such as occlusion,out-of-view and illumination variation during the tracking process of Staple algorithm can be detected in time.Target loss can be correctly detected by the proposed algorithm with a success rate of 100%.The proposed algorithm provides a reference for whether to perform target re-detection and target loss in practical engineering applications.