Algorithm Of Moving Target Detection Based On Multiple Information Merging In Hybird Color Space And Adaptive Earning Rate

Intelligent video surveillance is widely used in the field of military security, smart security,intelligent transportation and intelligent buildings, etc.,. It has developed into an important field ofcomputer vision research. Moving object detection is one of the key issues to be addressed inintelligent video surveillance, and their test results directly affect subsequent processing, hasbecome an important research focus. This paper on frame difference algorithm and backgroundsubtraction method in depth study and propose two improved algorithms.(1) The edge detection of digital image processing technology in this paper, color space,image denoising, morphological image processing and OpenCV platform respectively to itsprinciple, mathematical model and the corresponding performance analysis is introduced. Then,the development of video surveillance technology are introduced, moving target detection in theintelligent video surveillance is an important link and one of the problems to be solved. Movingtarget detection algorithm is the research object of this thesis mainly, in this paper, the movingtarget detection algorithm of the optical flow, interframe difference and background differencemethod basic principle, mathematical model and their characteristics are discussed and introducedin detail.(2) As the Gaussian mixture model background difference method, one of the most classic ofthe modeling method since put forward by the vast number of researchers attention. Aiming at theproblem of modal residual and ghosting for traditional Gaussian mixture model which has a fixedlearning rate, an improved algorithm of moving target detection by using adaptive learning rate ispresented. The variation characteristic of pixel on the image sequence and the model performanceof controlling parameters are analyzed, model of the learning process can be divided into initialformation and background maintenance updates in two stages. To adopt different learningstrategies in different stage, the initial formation stage, a bigger decreasing learning rate is adoptedto accelerate the background modeling. The Maintenance updates stage, according to the numberof the pixel matching and mismatching as feedback to adjust and implement model of adaptivelearning. Experimental results show that the improved algorithm can effectively improve theoriginal model of the slow convergence rate lead to problems in terms of background model updatenot in time, which can more accurately in detecting moving target. It is characterized by goodadaptability and robustness. (3) Aiming at the problem of leak detection, hollows and false targets in the traditional framedifference algorithm, an improved frame difference algorithm is presented. The effect of movingtarget detection is experimented and analyzed in several common color spaces, selecting excellentcolor channels to build hybrid color space CbVb*for moving target detection. In order to make fulluse of the frame difference correlation information, seven frame difference algorithm is presentedto calculate moving target frame difference in time domain according to the pixel changingcharacteristics of CbVb*space, edges of moving target in gradient domain is obtained by usingCanny operator with adaptive threshold, then frame difference in time domain and target edges ingradient domain are merged, and the merging information is carried on corrosion and expansionprocessing to get the final detection results. The experimental results show that the improvedalgorithm can more accurately detect moving targets, and has good robustness, adaptability andreal-time performance.