Font Size: a A A

Research Of Multi-camera Cooperative Target Localization And Tracking Techniques In Overlapping Field Of View

Posted on:2017-01-19Degree:MasterType:Thesis
Country:ChinaCandidate:J ZhangFull Text:PDF
GTID:2308330491950818Subject:Signal and Information Processing
Abstract/Summary:PDF Full Text Request
With the expansion of the scope of video surveillance and gradually complicated monitoring scenes, traditional video analysis techniques with pure 2-D image sequences are facing greater challenges. Therefore, the techniques using the relative target spatial coordinates which are recovered from 2-D images to assist video analysis become a hot research area. This thesis conducts a research on target spatial localization and tracking under multi-camera environment. The main contents and results of this work are as follows:1、Homograph estimation based on measurement data normalization. This thesis proposes an effective method that performing a normalization process before homograph estimation, which can improve the precision of homograph estimation. First,process the measured data method called condition adjustment to get more accurate GPS coordinates of identification points on scene plane; then design similarity transformation matrix to normalize the coordinates of the identification points on images and scene plane; Finally,we can calculate homograph with high precision through SVD.2、Fast people localization with multi-camera collaboration based on overlapping field of view. This paper presents an approach to achieve fast people localization with multi-camera collaboration which is sampling the people described by a rectangular box and giving a weight to each candidate foot points.In the process of producing the candidate foot points, the thesis samples the top edge of the rectangular box and projects the sampling points, and finally obtains the candidate foot points with multicamera geometric constraints.In the process of giving weights to the candidate foot points, the thesis get the center of gravity of foot points from the foreground information in the views in which the foot point is visible and then give the candidate foot points suitable weights according to the it’s distance from the center of gravity. Finally, compute the sum of the center of gravity and the weighted candidate foot points and take the average of the sum as the people localization result.3、The improved algorithm of target hand-off based on dividing line of view in overlapping field of view. This thesis puts forward a method of target hand-off by introducing the spatial location of target to assist the hand-off algorithm based on dividing line of view.This new method can achieve high reliability in seriously occluded scene. First, the dividing line of one camera in the field of view of its neighbor should be computed through homography;scend, execute the target hand-off algorithm based on dividing line of view to get the targets to be identified, and compute these targets’ spatial coordinate as well as reprojecting them back to the images; third, take the projecting point which is corresponding to the same spatial coordinate with the tracked target as the actual tracking target instead of the current tracking result, then, repeat the above process of localization and substitute until the targets in the current camera can be divided completely, finally, give the tracked target the label so that to finish the target hand-off.The thesis carries out the experiment and performance analysis of the above algorithm with simulated data and true real scenario. The results show that the homograph estimation based on measurement data normalization has better robustness. Fast people localization with multi-camera collaboration based on overlapping field of view can effectively handle the target localization in occluding and nooccluding scene. The improved algorithm of target hand-off based on dividing line of view in overlapping field of view can still accurately track the target even when there is occlusion by the dividing line of view between the tracked target and other target.The last part is a summary about the whole work, as well as a further discussion about the defects and possible improving direction of the research.
Keywords/Search Tags:target localization, target tracking, multi-camera cooperation, homograph geometric constraints, Field of View lines, target handoff
PDF Full Text Request
Related items