Moving Object Detection And Tracking

Moving target detection and tracking is one of the main problems of computervision research, it combines image processing、pattern recognition、automatic control、artificial intelligence、computers and other advanced technology in many areas,visualguidance in the military, video surveillance, medical diagnostics, intelligenttransportation and other aspects have a wide range of applications。Intelligent videosurveillance is an emerging field of computer vision and a concerned subject.It does notrequire human intervention, the use of computer vision analysis of the camera recordingimage sequences for automatic analysis, to achieve dynamic scenespositioning,identification and tracking of targets, and to determine the target behavior on this basis,thus completing the day-to-day management and responding in a timely manner whenabnormal situation occurs. With the continuous development of science and technology,stability、robustness of moving target detection and tracking system are put forward.Therefore, this research is important theoretical and practical value.This article mainly revolves around two aspects of moving target detection andmoving object tracking study, and introduce them under the concrete application. Interms of moving object detection, studied the static commonly used moving objectdetection algorithm, include flow method、the adjacent frame difference method、background difference method and so on. And then did the experimental comparisonabout the two background modeling method. Compared with the traditional gaussianmixture background modeling method, the new adaptive mixture gaussian distributionnumber K background modeling method has the not fixed pixel numbers of distribution,which can be adjusted according to actual situation, to describe the actual distribution ofpixel samples accurately; though an improved iterative algorithm, adaptive to completebackground extraction and update of model. Text used the video to320*240pixels, weprovide a target detection of moving pedestrian in interior space. Results show that thealgorithm is better than traditional Gaussian mixture background modeling algorithm, ithas a better ability to adapt, to update background have stronger timeliness andaccuracy.In terms of moving target tracking, now commonly used several kinds of algorithmare introduced, about tracking method based on the characteristics, tracking methodbased on3D, tracking method based on active contour and so on. Did the corresponding analysis and experiment for the color feature extraction and shape feature extraction,including RGB color space to HSV space conversion analysis.About shapecharacteristics,we mainly studied the graphics smoothing and edge detection,andexperimental analysis. Then I researched the particle filter algorithm mainly, andpresented an effective improved method, through clustering and hierarchical method,the particle space is divided into layers of space, for the weight of different space usingparticle size, to ensure the diversity of the particles and effectiveness, and avoid thewaste of the particles. In order to verify the effectiveness of the algorithm,we use two320*240pixels video to text. Experiment show that the improved algorithm is betterthan the traditional particle filter. Through the simulation, we know that the improvedmethod of the tracking error is less than half that of the original algorithm, eachsimulation time stability is strengthened, and the tracking precision is also improved.