改进KCF的尺度自适应目标跟踪算法研究

Research on Scale-Adaptive Object Tracking Algorithm by Improving KCF

针对KCF跟踪算法在目标跟踪过程中存在目标尺度变化时检测精度低、目标遮挡时跟踪容易丢失等问题,提出了SMAKCF(Scale-Adaptive Multiple-Feature Anti-Occlusion KCF)跟踪算法,该算法同时优化了KCF算法中的尺度响应、特征选择及模板更新策略,融合HOG特征及CN特征,加入尺度估计滤波器并利用APCE判据改进位置滤波器的更新方式,同时引入了一个检测模块对不可靠跟踪结果进行重检测。在Visual Tracker Benchmark的50个测试视频序列上进行实验来评估算法的性能,实验表明,SMAKCF算法能够有效地解决目标的尺度变化及遮挡问题,提高跟踪算法在长时目标跟踪过程中的性能。

The SMAKCF(Scale-Adaptive Multiple-Feature Anti-Occlusion KCF)is proposed to solve the problem that the KCF algorithm can not adapt to the object scale and occlusion when tracking an object.The SMAKCF algorithm optimizes simultane-ously several problems including scale response,feature extraction,and update strategy.To be specific,a fusion feature is put for-ward combing HOG features and CN features efficiently,then a scale estimation filter is added and the APCE criterion is introduced to improve the updating method of the position estimation filter.Besides,an extra detection module is designed for re-detecting the object which is unreliably detected.Experiments are conducted on 50 test video sequences of Benchmark to evaluate the algorithm performance.It is indicated that SMAKCF algorithm can overcome difficulty in the scale change and occlusion of the object,the tracking ability in the long-term object tracking process is enhanced significantly.