Research On Video Target Compressive Tracking Algorithm With Fusion Depth Information

Video target tracking is to make the computer imitate the human visual perception mechanism to obtain information of the interest region in continuous video frames such as the external outline and motion trajectory.In recent years,target tracking has been a hot research topic in the field of machine vision and pattern recognition.It has important applications in video surveillance,human-computer interaction,intelligent transportation,medical treatment,military and other fields.Affected by factors such as complex background,changes in light,target deformation and occlusion,improving the robustness of the target tracking algorithm is still a challenging research topic.In order to build a robust tracking algorithm with high accuracy and robustness,this paper applies compressive sensing theory and information fusion theory to research the appearance model feature,depth information and gray information.Multi-feature weighted appearance model and classifier construction and update mechanism of tracking algorithm was studied and proposed two target tracking improvement algorithms.The main work of this article is as follows:1.A weighted appearance model with fusing depth information of video target compressive tracking algorithm was proposed,which effectively improves the accuracy of the target tracking in complex environments.Aiming at the problem that the algorithm has weak ability with a single feature under the environment of encountering changes in light illumination,similar background,pose change and occlusion,a multi-feature appearance model was built and it complemented the feature representation of the target appearance model.By fusing multiple features,the ability of describing the target is stronger.The coarse-to-fine search strategy and the compressive feature of the original feature information were used to ensure the real-time performance of the algorithm.At the same time,the Mahalanobis distance was assigned to the weights of classifiers that improve the weights of the effective weak classifiers.The proposed algorithm can effectively deal with complex environmental changes and improve overall tracking performance.2.A blocking anti-occlusion video target tracking algorithm with depth information fusion was proposed to effectively deal with the application scenarios such as background similarity,target attitude change and partial occlusion.Using sub-block haar-like features of the color image and the sub-block depth histogram features of the corresponding depth image to establish a joint model of target appearance to effectively improve the target description capability;the target tracking was considered as a binary through the weighted naive Bayes classifier of the Bahrain coefficients.The classification problem was used to determine the target tracking result.The target occlusion detection mechanism was enhanced by setting the sub-block maximum classification score of color image and sub-block similarity threshold of the depth image.The proposed algorithm can effectively reduce the probability of extracting false positive and negative sample information and improve the robustness of target tracking.3.In this paper,quantitative analysis of the two improved algorithms was performed in the PTB data set video sequence.Compared with the current popular tracking algorithms,including meanshift algorithm,real-time compression tracking(CT)algorithm and fast compression tracking(FCT)algorithm.Experimental results show that the first algorithm proposed in this paper improves accuracy and robustness while maintaining real-time performance.Although the second algorithm has lower tracking speed than the CT algorithm,it performs well in terms of target rotation and occlusion.