Unsupervised mine personnel tracking based on attention mechanism

Object tracking is a challenging computer vision task, which plays an important role in the fields of intelligent transportation, human-computer interaction, and video surveillance.There have been many tracking algorithms with superior performance, but it is still difficult to achieve good tracking results in coal mine scenes.It mainly faces challenges such as severe occlusion, more background interference, and more people underground, which seriously affects the performance of tracking.Secondly, there are challenges such as the small number of samples in the data set and the lack of uniform labeling.This paper designs an unsupervised method to train the target tracking model in the coal mine scene, where the mine video data set is imperfect, the image quality is low, and the lack of unified annotations.The combination of discriminative correlation filtering and twin networks takes the advantages of correlation filtering and Siamese networks, and constructs a lightweight end-to-end target tracking network model.Specifically, the forward tracking of the target and the backward verification of the target position information of the previous frame are adopted to realize the target tracking process of the unsupervised model.According to the problems of serious occlusion, more background interference, and close arrangement of dense targets in the mine environment, this paper proposes a method of using the attention mechanism to extract important information in video images.The channel attention mechanism and spatial attention mechanism are added to the backbone network structure of the model, which extracts the focus of important attention.The backbone network of the model uses lightweight AlexNet network to solve the problems of limited storage and computing resources of mobile platforms in the coal mine environment.Experimental results show that the tracking model proposed in this paper is suitable for the personnel tracking in the complex environment of coal mines, and has a good detection effect.