Multi-feature Fusion And Adaptive Scale Moving Target Tracking Method

Target tracking is a hot topic of research in the field of computer vision.At present,many scholars have introduced correlation filters into the target tracking framework and achieved more results because of the significant effects of the correlation filter algorithms,but at the same time,there are still many problems in the target tracking field,such as occlusion problems,target scale changes and background information.Interference,etc.,these factors will affect the target tracking effect.Improving robustness of target tracking,this paper studies and analyzes the issues of feature selection,filter template update and target scale change.The main work is as follows:(1)Propose the strategy of adaptive selection feature,extract the direction gradient feature and color feature of the target separately,and train the respective filter models.Then,according to the peak sidelobe ratio of different characteristic response graphs in the detection phase,feature fusion using adaptive weighting algorithm,this method works well when the target is occluded.(2)Propose an adaptive update filter model strategy,first set a threshold,and then judge whether the target is occluded according to this,if the occlusion does not update the model,otherwise the filter model needs to be updated,this method improves the filter The validity of the template.(3)The scale estimation method is introduced to update the scale strategy of the target in real time.First extract the sample direction gradient feature and construct a scale pyramid for training the filter,and then get the maximum response value,that is,estimate the target scale.This method adaptively changes the size of the target tracking frame and improves the tracking accuracy.(4)According to the above research,an adaptive scale-changing target tracking method based on multi-feature fusion is proposed,and the filter model is selectively updated.The improved algorithm proposed in this paper is compared and analyzed in the OTB test data set.The results show that the proposed algorithm has higher tracking accuracy and real-time performance under occlusion and scale changes compared with other algorithms.