Study On Target Tracking Based On Compressive Sensing

As a new sampling theory, Compressive sensing can perfect reconstruct the original signal with non-linear reconstruction algorithm in far less than the Nyquist frequency sampling rate by utilizing the prior information that signal always can be sparse represented in certain dictionary and the incoherence measurements. Compressive sensing theory is not only a great breakthrough of traditional sampling theory, but also provides new ideas and methods for the research of other scientific fields. By Using Sampling theory of compressed sensing which can discard target feature redundant information, the traditional target tracking method can effectively improve the timeliness and accuracy of tracking.In order to improve the stability of target tracking algorithms in different complex environments, this article conducts systematic and in-depth research on target tracking algorithm based on compressed sensing.The main work of this article is summarized of the following aspects:(1) The theoretical framework of compressed sensing systems theory is systematically researched, and the solutions of the problems of Compressive sensing are discussed. And the theoretical foundation of Target Tracking based on compressed sensing is given.(2) For the traditional particle filter tracking algorithm that exists in real-time and accuracy problem, a particle filter tracking method based on Compressive sensing is proposed. By studying the theory and particle filter tracking method, the target model and state estimation process of particle filter tracking is improved using Compressive Sensing theory; Experimental result verify that algorithm based on Compressive sensing can achieve real-time tracking and robustness.(3) As traditional tracking algorithm based on online learning is unstable to track target and easily lead to drift or tracking failure when the target quickly move, the texture or the environment get seriously changed, an improved target tracking algorithm based on compressive sensing is proposed. By extracting the gray and texture features using compressed sensing, and automatically calculate the features’ weight according to the classification results due to the stability of different features. Results of tests on variant video sequences show that the proposed algorithm is superior to the other tracking algorithm based on compressive sensing in terms of tracking accuracy and real-time performance.