Research On Object Tracking Based On Deep Learning And Correlation Filters

Visual object tracking algorithms have been widely studied by researchers.In recent years,many excellent algorithms have been proposed,but visual tracking still has great challenges,such as target scale change,target fast moving,illumination change,target deformation,background clutter,target occlusion and many other interference situations.On the basis of analyzing the existing target tracking algorithms,we study the object tracking under interference conditions based on correlation filtering and deep learning theory.The specific work is as follows:For the correlation filtering tracking algorithm is not robust enough and cannot adapt to scale changes,target occlusion and other complex interferences.We introduce a correlation filtering trcking algorithm based on superpixel and multifeature fusion.First,superpixel segmentation and clustering are performed for the target and its surrounding environment in the initial frame.On this basis,the histogram of gradient(HOG)and HSI color features of the target sub-block are extracted to interact with their respective position filters.Accordingly,we independently train a scale filter with a multiscale pyramid constructed at the estimated target location to estimated the object scale.Lastly,we introduce an occlusion criterion for determining whether to update the model or not.Combining depth features and related filtering,we proposes an adaptive target tracking method based on channel attention mechanism and layered convolution feature fusion.First,the VGG-M network is used to extract multi-layer features to characterize target semantic information.The channel attention module is introduced to calculate the weight of each channel to ensure the effectiveness of the channel so that our method can adapt to the taiget deformation.Then,online training is perfomed on the correlation filter corresponding to each layer feature,and the response graph is calculated separately.Lastly,we design a response adaptive fusion strategy to further improve the performance of the tracker.In this study,we use OTB100 and VOT2017 test sequences.A large number of experiments were designed to verify and analyze the performance of the algorithm in this chapter from multiple perspectives.Experimental results show that the target tracking algorithm proposed in this paper can still achieve accurate and robust target tracking in the face of multiple interference situations.