Moving Object Automatic Detection, Extraction And Tracking

This article is in the camera fixed, and to operate object detection,extraction and rracking in the background is relatively static conditions. Moving target detection is to determine whether the target crash into monitored area,testing whether there is relative motion object to backgrand.Object detection is operated after objected is detected and to completely separated object from background. Target tracking is to record trajectory of target for the future have a higher level of classification,providing data to support behavior understanding. From target detecton,target extraction,to usage of target extraction as a mean shift algorithm for tracking the initial search window that required to complete the automatic tracking,and by setting the relative changment in the amount of similarity measure to accommodate the target size,shape changment,in whole process without human intervention to reach intelligent detection and tracking results.In this paper,first part is to introduced the research backgroud and significance,status of research in this field,contents of this research and a brief chapter arranged.Before making detailed descripition of object detection and extraction,there is a relative prior knowledge introduction,including mathematical morphology,shadow elimination,connected component labeling,ects.In moving tagert detection and extraction,it study the most commonly three methods:it give a detailed description of frame difference method for the specific processes and procedures,and using Ostu method to select the partitioning threshold; In the backgroud subtraction,there is a brief introduction to pixel gray classification algorithm,Surendra algorithm and means methos,all that is used to construct the backgroud image.Finally,it introduced self-backgroud model adaptation that is used to detect,extract the foreground region,and a detailed description of the single Gaussian model and Gaussian mixtured model and updated strategy parameter in the model,to determine the moving area by judging whether the pixel color value to meet the Gaussian model. In target tracking,first it introduced the basic principles of mean shift algorithm. However, the traditional mean-shift algorithm requires manual calibration of the target area, this improved mean shift algorithm uses the results of target extraction as the initial data of target tracking, in order to achieve target tracking under unattended conditions. and to provide the initial search window with combination of object extraction method. In tracking process, by increasing the relative changment in amount of similarity measure to decide whether to re-obtain the research window to solve the veracity problem when target size and shape changement have some changement. The experiement proved the algorithm achieved good results.