Compression-based Moving Target Detection Methods Perception

Compressive sensing is an emerging signal sampling theory, different from the traditional Nyquist signal sampling theorem, it provides a framework for image recovery using sub-Nyquist sampling rates. With further research of the theory, it will make a deep influence on the traditional signal sampling. In recent years, the compressive sensing theory has been applied to various fields of signal and image processing.Moving target detection is an important part of the computer vision research and also difficult. It detects and extracts the moving target it in the dynamic picture sequence, in the process of moving target detection, we must be faced a large number of sample data of the image sequence. For this reason, it is difficult to the system that high sampling rate is difficult to achieve, and the communication bandwidth and capacity of storage is limited. By the advantage of lower sampling rate and combining the compressing with sampling, Compressive sensing overcomes the shortcomings of sampling, transferring and storaging of traditional image. it is a good way to apply the compressive sensing theory in these systems.In the paper, we propose a moving target detection method based on compressive sensing. By deep study of the theory, we cast the background subtraction problem as a sparse image recovery problem, describe a method of constructing the sparse foreground image directly in compressive sensing domain and detect the moving target in compressive sensing domain with Kullback-Leibler divergence. Compared with the traditional method of moving target detection with background subtraction, we do the detection and update the background model in a lower dimensional compressive sensing domain. When the sampling rate of image data obtained by compressive sensing is low, the performance of the image constructed by these sampling data is poor, we can’t accomplish the moving target detection by these constructed images in the spatial domain accurately. Due to the sparsity of the foreground image, the performance of foreground image is better than the image obtained by traditional method if we accomplish the moving target detection in the compressive sensing domain. Experiments show that the algorithm has a better performance with the image data obtained by lower sampling rate in compressive sensing, and can robust against the change ofbackground.