The Research Of Motion Objects’ Detection And Tracking Among Multiple Cameras’ Scenes

With the high speed development of economy and the rising of living level of the people,they began to pay more and more attention to the security problems in their living and working.The problem, the government and scientists are faced with, how to ensure the living and propertysafety of the people effectively is becoming urgently needed to be solved. There are two primerydifficulties in the intelligent monitoring system of multiple cameras: the motion objects detectionand tracking in the single camera sence, and the motion objects matching and continue trackingamong the multiple cameras sences.The research background is mainly on intelligent surveillance system. This paper mainlyresearchs the objects’ tracking in the multiple cameras’situation, which incluldes the condition ofoverlapping and non-overlapping of the field of views. The goal of the topic is analysising videosequences collected by the multiple cameras,extracting the feature information intelligently andtracking the object continuely.The paper’s main work includes the following several aspects:(1) The detection of the motion objects.Background model is built firstly, then the motionobjects are extracted by the means of combing the background subtraction and frame subtraction.Shadow is wiped out by the feature that it has similar chromaticity but lower brightness thanthose of the same pixel in the background image.At last, morphology methods and median filtersare used to delete noise and make up cavitas to acquire a clean and accuracy object area.(2) The approach that matches of objects in the situation of overlapping field of views isproposed.This method uses the projective invariants to get the field of view lines between theadjacent cameras.It only needs to be computed once.The method is simple, convenient andaccuracy. In the process of tracking,when a motion object leaves the field of view of one cameraand enters the field of view of another camera,compute the projective location in the secondcamera,make the Euclidean distance between two objects’ locations as the match criterion.If onlyone object’s location to the projective place’s distance is less than the threshold,they are the sameobject;If none of the objects’ locations to the projective place’s distance are less than thethreshold,the matches are fall and they are the different objects;If more than one objects’locations to the projective place’s distance are smaller than the threshold,fusion the SIFT featureto match and the best match means the same object.Research and experimental results show thatthe objects matching based on the field of view lines is accurate and get good instantaneity. (3) The approach that matches of objects in the situation of non-overlapping field of views isproposed.Extracting the object’s major color spectrum histogram and global color model whenhe enters the field of view of camera1.If the object leaves the field of view of camera1andenters the field of view of camera2,extracte the major color spectrum histograms of all theobjects that detected in camera2.Match all those histograms of camera2with the histogram ofcamera1according to the match rule.The best match is judged as the same object.If two objects’appearance features are too similar,then fuse the global color model. Experimental results showthat the fused method is more robust and makes better matches.(4) Fusioning the Camshift algorithm and Kalman filter as the objects’ tracking method.UseKalman filter to predete the location of the motion object where it may be appeared and therange of searching window in the next frame,then do Camshift algorithm in the nearby of thepredeted location. Experimental results show that this method can increase the degree ofaccurancy effectively and decrease the degree of lose.Finally, the summary of this thesis is made and the future study direction on this field isgiven as well.