Weighted sparse cooperative model-based target tracking method

Abstract

The invention discloses a weighted sparse cooperative model-based target tracking method. The method comprises the following steps of: 1, selecting a tracking target from the first frame of image data of a video sequence; 2, initializing any frame f=1; 3, processing the fth frame of image data in the video sequence by utilizing a particle filter algorithm so as to obtain a plurality of target candidate boxes; 4, processing the target candidate boxes in the fth frame of image data by utilizing a weighted sparse representation-based discriminant algorithm so as to obtain discriminant scores of the target candidate boxes; processing the target candidate boxes in the fth frame of image data by utilizing a weighted sparse representation-based generative algorithm so as to obtain generative scores of the target candidate boxes; and 6, obtaining final scores of the candidate boxes, comparing the scores of all the candidate boxes to find a candidate box corresponding to a maximum value, and taking the found candidate box as a tracking result. According to the method, real-time movement estimation and positioning can be carried out on moving targets in a video sequence, so that the stable tracking for the moving targets is realized.

Description

technical field [0001] The invention belongs to the field of video monitoring, and proposes a weighted sparse cooperative model-based target tracking method to stably track a moving target in a video sequence. Background technique [0002] In recent years, object tracking has played a very important role in the field of computer vision, and it is also a research hotspot. With the continuous development of target tracking technology, it plays a vital role in a variety of practical applications, such as: video surveillance, human-computer interaction, behavior analysis, virtual reality, automatic control systems, etc. Scenes. Therefore, a variety of target tracking methods have emerged, mainly based on generative models, and also based on discriminative models. [0003] The generative model is based on the target detection, after modeling the appearance of the foreground target, according to a certain tracking strategy to estimate the optimal position of the tracking target,...

AI Technical Summary

Problems solved by technology

However, the correlation between test samples and dictionary atoms (that is, training samples) is not considered in both the sparse production-based model and the sparse discriminant-based model, which will make the obtained sparse coding coefficients inaccurate, thus affecting tracking accuracy

[0004] In order to improve the tracking accuracy of the moving target, the researchers adopted the tracking method of the sparse representation hybrid model, which comprehensively utilizes the global template and the local expression, and can efficiently deal with the apparent change of the target. However, in the process of solving the coefficient encoding coefficient of the hybrid model, the obtained The coefficient sparse coding cannot be sparse enough, so the positioning of the moving target is not accurate enough, and the long-term stable tracking of the target cannot be achieved

Images