The Research Of Background Subtraction Based On Compressive Sensing

Compressive Sensing is a groundbreaking theory in signal and informationtheory was formally proposed in2006. In the theory signal sampling and compressioncan be performed simultaneously, using the calculation method for solving theoptimization, based on only a small sampling signal, reconstruction algorithm can beused to restore a more accurate signal that has sparse or compressible nature. In thefield of signal processing, applied mathematics, electronics and communicationsengineering, computer science, compressive sensing has attracted wide attention. Inthe field of computer vision research, especially intelligent video surveillance system,moving objects detection algorithm is a very important research topic. Compared withthat has been put into practical application of various moving target detectionalgorithm, a background subtraction algorithm are widely used. In view of theadvantages of compressed sensing theory, consider the intelligent video surveillance,in order to achieve complex multidimensional signal processing needs, reduceoperational complexity, reduce hardware costs, and many other issues, the goal of thisthesis is the background subtraction method based on compressive sensing.This thesis mainly research work are: firstly introduces the basics of compressedsensing, theoretical framework, the implementation process, domestic and foreignresearch status and related applications. Then focuses on the signal sparserepresentation, measurement matrix, reconstruction algorithm, including theapplication of the premise, selection factors, and optimization algorithms.Following focuses on how to use compressed sensing theory framework to buildbackground subtraction method to achieve the target detection algorithm. The basicidea of the algorithm is: a background image in the known case, because of thedifferential image is only small proportion compared to the background image, it’smeasurement values satisfy sparsity, using the ultra-complete dictionary sparserepresentation. The measured value of image obtained by the measurement of theimproved Hadamar matrix. In compressed sensing domain model to establish thebackground subtraction, The measured value of difference image can be obtained byusing the difference between the test image and background image, the differenceimage compression observations completed. Using adaptive background updating method to update the background image, eliminating the interference of variousbackground changes to ensure the accuracy of the moving target detection. Throughsimulation experiments, which can verify the algorithm can be efficiently sampledand can obtain a more accurate reconstruction of images is feasible. While thealgorithms exist some shortcomings, such as the signal measurement time is longer,the measurement matrix structure determines the scope of its applications are limited,the overall performance of the practical application of the algorithm needs to befurther improved.