融入运动信息和模型自适应的相关滤波跟踪

Correlation Filter Tracking Incorporating Motion Information and Model Adaptive

为解决余弦窗的影响和复杂场景中的目标遮挡问题,提出了一种融入运动信息和模型自适应的相关滤波跟踪算法。采用HOG特征和颜色直方图特征互补结合的框架,引入卡尔曼滤波和上下文感知滤波器,可以解决余弦窗的影响;引入一种高置信度检测方法和一种新的模型自适应更新方法,可以解决目标遮挡的问题。将提出的算法在OTB2015测试集与其他6种相关滤波类算法进行比较,实验结果表明,该算法精确度和成功率分别为0.821和0.615。相对于StapleCA算法,精确度提升了1.3%,成功率提升了2.8%,同时,算法速度为54.34帧/s,满足实际工程实时性要求。

In order to solve the problem of cosine window influence and target occlusion in complex scenes,a correlation filtering tracking algorithm incorporating motion information and model adaptation is proposed.Adopting the framework of complementary combination of HOG features and color histogram features,introducing Kalman filter and contextaware filter are introduced to solve the influence of cosine window;introducing a highconfidence detection method and a new model adaptive update method are introduced to solve the problem of target occlusion.The proposed algorithm is compared with other 6 related filtering algorithms in the OTB2015 test set.Experimental results show that the accuracy and success rate of the algorithm are 0.821 and 0.615,respectively.Compared with the StapleCA algorithm,the accuracy has increased by 1.3%,the success rate has increased by 2.8%,At the same time,the algorithm speed is 54.34 frames/s,which meets the realtime requirements of actual engineering.