Research And Implementation Of Moving Object Detection Algorithm Based On Background Modeling

With the rapid development of computer vision technology,the demand of efficiently and accuratly extraction of moving object from video is increasing.The technology of moving object detection has important meaning in artificial intelligence,human-computer interaction and virtual reality.However,the video is often disturbed by various natural environments noise and man-made noise that makes the process of obtain a clear and significant moving object facing great challenges.There are various motion object detection algorithms,but some problems still need to be resolved.For example,if the background model is poorly designed,the algorithm will cause many misjudgments,and it will also affect the subsequent update of the sample set and spread misjudgment;in another way,if the front-background is not correctly segmented,some of the image information will be lost and it will introduce some non-existent information to the results.These problems will affect the quality of the algorithms.This paper will improve these problems.In order to avoid the introduction of extraneous noise and other information before modeling,this paper first selects the minimum value of the light intensity in the image before the background modeling and use the dark channel image reduce the interference of noise.Then evaluates the difference between gray image and dark channel image.Then,the dark channel image is smoothed by filtering of itself,so that the preprocessed dark channel image can reduces the noise interference and improve the quality of the detected moving object.According to the characteristic of the dark channel image after the self-directed filtering maintains the texture consistency with the original image and reduces the noise,this paper propose a robust algorithm of moving target detection based on background modeling.The algorithm avoid the situation of that the background model is easy to misjudgment when the noise interferes.In order to reduce the interference caused by the uncertain motion of the moving object,this method use a hybird update strategy that combines conservative updating and counting updating to smooth the sample set of the background model,after obtaining the exact type of pixel in the pixel classification stage,a random diffusion background model updating method is used in this paper.At the same time,the background model of the pixel's neighborhood are also updated randomly,thereby improving the accuracy of the background model.The proposed method not only improve the accuracy of sample classification,but also can get better quality image of the moving object.The image is susceptible to noise pollution,and the object detection algorithm is usually sensitive to noise,so it is necessary to study the object detection of video images with noise.In this paper,a robust principal component analysis method combining adaptive sample and threshold updating is proposed.This method decomposes the background and foreground by performing singular value decomposition from the perspective of considering the moving object as the sparse noise of the image.This method not only can adapt well to various scenes,but also can accurately decompose moving object in complex scenes.In order to reduce the time consumption caused by multiple iterations,an inaccurate approximation method is used to speed up the iteration in the process of the decomposition of foreground and background.By performing the difference analysis on the adjacent frames to adjust the size of the sample set adaptive that can reduce the number of samples used in the decomposition proces and improve the speed of the algorithm.The adaptive threshold segmentation method is used to classify the background of the pixels,which improve the accuracy of the segmentation and get clearer and complete object by combining the spatiotemporal information of the pixels.The proposed method not only improve the speed,but also achieves a higher quality of the object image.In summary,based on the study of moving object detection,this paper propose two improved algorithms for moving object detection.The algorithms are improved the speed and accuracy by using dark channel prior and robust principal component analysis.The proposed methods improves the accuracy of the algorithm with effectiveness of moving object detection.