Research On Algorithm Moving Object Detection And Tracking Inintelligent Video Surveillance

In computer vision filed, motion target detection and tracking is one of the most importanttechnologies for intelligent video surveillance.Moving targets in image sequence detection, extraction,location, tracking and recognition can be completed without manual intervention. Due to a largenumber of visions informationsare included in the motion, so the motion target detection and trackingis one of the important research field. Motion target detection and tracking is the main research objectin the thesis.First of all, the paper introduces the video monitoring technology development and research statusat home and abroad, as well as the application fields of video surveillance technology, technicaldifficulties and development direction.Second, the paper presents a video image preprocessing methods. Since video surveillance movingtarget detection and tracking is based on the extraction and analysis of image sequences over a regionof interest, therefore, a variety of image segmentation in image preprocessingimage filtering, imageprocessing techniques in this paper are briefly introduced.Third, the paper proposes an improved an parameter update method for Gaussian mixture model.For commonly used method for detecting and tracking moving targets, the paper made a detailedanalysis, including the classical algorithm of moving target detection method of optical flow field,frame difference method, background difference method, etc. In the moving object detection,according to the different characteristics of the mean and the variance, use different rates of learning.This improved learning mechanism allows quick and accurate convergence of the mean, varianceconverges faster and more stable, experimental results show that the proposed algorithm can achievebetter foreground detection results.Fourth, based on the classical mean shift algorithm, the fusion of the spatial location characteristicsand the integration of mean shift is proposed. The method introduces the spatial information to predictthemotion target range in the next frame, According to the target at any moment of physicalmovement in the prediction of target bearing1, and according to the classical mean shift algorithm oftarget bearing2, fusion this two target bearing, we can forecast the final actual target bearing.Experiments show that the algorithm can effectively improve precision of the target matching andtracking accuracy. Finally, for target tracking in a complicated environment problem, the paper conducted in-depthresearch, a target tracking method based on adaptive template updating is proposed. This methodcombined of the extraction of scale invariant feature and the fast template matching, the simulationexperiments show that the proposed algorithm can better solve the problem such as target trackingchanges in light and shade, and has good real-time performance.