一种改进的背景感知相关滤波跟踪算法

A Modified Correlation Filter Tracking Algorithm for Background Perception

针对视觉目标跟踪复杂环境中的遮挡、运动模糊、形变以及背景干扰等问题,提出了一种改进的背景感知相关滤波跟踪算法,通过增加提取灰度特征(Histogram of Oriented Gradient,HOG)及颜色特征(Color Names,CN),提升了目标定位精度,并在此基础上采用多峰检测的模型更新策略抑制相似特征,解决了模型漂移问题,最后利用训练尺度滤波器进行尺度估计得到了目标估计大小。实验及仿真结果表明:所提算法在OTB-100所有数据集下,较背景感知相关滤波(Background-Aware Correlation Filters,BACF)算法在准确率和成功率上分别提高了1.3%和1.4%,中心位置误差平均降低67.23个像素点,重叠率提高16.1%,平均运算帧率可达到10.09帧/s,有效提升了算法性能,同时解决了跟踪中遮挡、运动模糊、形变以及背景干扰等问题,具有较高的理论价值和工程应用价值。

Aiming at the problems of occlusion,motion blur,deformation,background interference and other problems in the complex environment of visual target tracking,a modified background perception-related filtering tracking algorithm is proposed.By adding HOG(Histogram of Oriented Gradient)and CN(Color Names)features,the accuracy of target positioning is improved.On this basis,the model update strategy of multi-peak detection is used to suppress similar features and solve the problem of model drift.Finally,the target estimation size is obtained by using the training scale filter for scale estimation.Experimental and simulation results indicate that the proposed algorithm is 1.3%and 1.4%higher than the BACF(Background-Aware Correlation Filters)algorithm in all data sets of OTB-100,and the center position error is reduced by 67.23 on average.The overlap rate is increased by 16.1%,and the average calculation frame rate can reach 10.09 frames/s,and the algorithm performance is effectively improved.At the same time,it solves the problems of occlusion,motion blur,deformation,background interference,etc.in tracking,which has high theoretical value and engineering application value.