基于非经典感受野抑制的TLD目标跟踪方法

TLD Target Tracking Method Based on Non-Classical Receptive Field Suppression

为了提高目标遮挡或存在虚假目标时系统的跟踪精度,提出了一种基于非经典感受野模型(NCRF)的跟踪-学习-检测(TLD)跟踪算法。根据TLD算法的目标跟踪框架,主要研究了Lucas-Kanade跟踪算法、级联分类器和P-N学习;通过分析NCRF的2种抑制特性,即朝向选择性抑制和非朝向选择性抑制,建立了基于NCRF抑制特性的轮廓检测模型;为了提高目标跟踪精度,结合TLD算法和NCRF抑制轮廓检测算法(TLD+NCRF)对目标进行跟踪。通过试验计算与分析,视频跟踪结果表明,TLD+NCRF算法的跟踪效果优于TLD算法。

To improve the tracking accuracy in case of shielded target or false target,a tracking-learning-detecting(TLD)tracking algorithm based on non-classical receptive filed(NCRF)model is proposed in this paper.Firstly,the Lucas-Kanade tracking algorithm,the cascade classifier,and the P-N learning were studied based on the TLD target tracking framework;Secondly,the two suppression characteristics,namely,orientation selective inhibition and non-orientation selective inhibition,were analyzed,based on which,a contour detection model was built.Finally,to improve the tracking accuracy,the study exercised the target tracking by combining TLD algorithm and NCRF contour detection model.The results from testing and analysis show that the TLD+ NCRF algorithm has performed better than the TLD algorithm.