| Intelligent Video Surveillance System is now used more widely, and intelligent video analysis algorithm has arose great interest around researchers. One of the valuable topics is object tracking by cooperated dual cameras. In practical use, dual cameras cooperated object tracking system can enlarge field of vision, precisely fix the position of object, solve the problem that the object is covered, thus has a good practical foreground. The thesis studied some key problem under dual cameras cooperated object tracking system by designing an experiment system, and analyse the performance of the algorithms proposed. The thesis aim at solving the problems including object detecting and tracking, object handing over, and situation of scene with superposition or distance.Object detecting and tracking is the first step of intelligent video surveillance. The thesis use inter-frame difference method to detect the moving object. We also introduce texture detection and contour detection to make the object detection more precise. We use CamShift algorithm to accomplish the object tracking.Cameras must communicate(exchange data) to ensure that the object could be properly handed over. The more information exchanged, the better the result would be, but cost of communication increased,while the less information exchanged, the worse the result would be.This thesis uses outter rectangle of object and chroma feature as exchanged information between dual cameras, and uses trace of object to achieve handing over it.In some case, when two scenes overlapped, mistake might occur while using trace for handing over.When two scenes has a distance between them, we need estimate to make the result more precise.We use SIFT to detect a boundary for the overlapped case, such that trace could be properly handed over. We use moving speed of object to estimate the time and position when it appear. Finally, we discuss the case of multi-objects.We make a verification and analysis for our algorithms by experiment. |
PDF Full Text Request