Object Tracking Using Mean Shift Based On Four Channel Non-Separable Wavelets

As an important research direction in computer vision,moving object detection and tracking has become a research focus,for it refers to image processing,artificial intelligence,automatic control,pattern recognition and so on.At the same time,the applications of this study,like military guidance and interception,video surveillance,intelligent traffic,medical diagnosis,have contributed to its heat.Tracking the target accurately in a variety of complex environment,such as cover,illumination change,deformation and so on,makes this hot issue a difficult problem.A tracking algorithm which can take into account the performance and efficiency of tracking and ensure strong adaptability has become a difficult and difficult point in the study.To solve this problem,a target detection algorithm and a tracking algorithm based on four channel non separable wavelets are proposed in this paper.In target detection,this paper constructs a four channel non separable wavelet filter which is suitable for target detection.At the same time,the isotropic characteristics of the four channel non separable wavelet filters is used to decompose the gray image of the original image.Image fusion is performed using three high frequency sub images to enhance the difference between the edge part and the other parts in high frequency subgraphs.The threshold of the high frequency fusion subgraph is taken to obtain the edge of the target in the image.The image segmentation is carried out by the morphological processing results of the edge graph,and the target and the background are separated from the image.In target tracking,the high frequency subgraph and the low frequency subgraph of the effective component in the RGB space are used as fusion features to track in this paper.Every several frames,the high frequency sub bands and low frequency sub image of the three color channels are used to track using mean shift tracking separately.According to the tracking results,the similarity of the corresponding position is calculated.And the color which has the highest similarity is chosen as the only color component to track in the next few frames.In order to adapt to different situations,we decompose each color component by non-separable wavelet transform and used adaptable weights between the two subgraphs.According to the result of image segmentation,the smallest outer rectangle of the target area is taken as the size of tracking box in the next frame.According to the similarity ratio of the high frequency and low frequency subgraphs in the tracking results,the weights of the features are determined.In the experimental verification and comparison part,this paper verifies the adaptability of the optimized algorithm under different conditions,and compares it with several tracking algorithms.The results show that the proposed method is real time and accuracy when the target appearance changes and the background changes.