Compressed Sensing And Its Appliation In Target Tracking

The advantage of Compressed Sensing(CS) relies on high rate of sampling. The signals sampled according to CS theory can be reconstructed in terms of probability of one. Accordingly, CS has had wide application on image processing and computer vision, target tracking especially. As for the importance of real-time, this paper improve the extraction of feature of object:Firstly, as for high complexity, CS reconstruction algorithm can not be applied to target tracking, the compressed random sampling is applied to target fern feature extraction. Fern random characteristics of the target is different from the set consisting of pixels, each element of the set as the difference between the pixel by pixel grayscale values, the collection complexity of the algorithm has been effectively reduced. In order to meet real-time requirements, we use random measurement matrix of random sampling fern compression, according to the nature of CS RIP theory, stochastic characteristics of the sampled fern can well meet the needs of target recognition and tracking.Secondly, we proposed a "sparse subspace particle filter", the compressed algorithm and sparse sampling subspace representation applied to the target tracking. First, the target compressive tracking sampling, in accordance with the principles of compressed sensing signal is sampled to maintain the original amount of information at a higher probability, provide enough information to ensure the target tracking. Next, the robust principal component analysis(RPCA) method to extract the compressed sparse sampling target the main ingredient. In this paper, creative usage of ?1- ?2 norm to solve problems in order to obtain the target RPCA sparse principal component. At the same time, the process for tracking particle filter in the theoretical framework: specifically refers to the use of continuous time series over the goal under the state estimate the current state of the target time. In the target sparse subspace where the update process, the candidate target subspace and similarity update subspace.