Vehicle Detection And Tracking With Multi-Camera

With the construction and development of Ping'an City and Intelligent City,dotted video surveillance is everywhere.Therefore,the research of continuous target tracking and motion state analysis under multiple cameras environment has raise high awareness,which has become video analysis' hot research content.This thesis is aiming at the relevant problems that encountered in the multiple cameras intelligent surveillance system.It makes the thorough research about the target tracking,target handover and camera cooperative control technology in the multiple cameras environment after the current existing research's studying and analyzing.The specific contents are as following:1.It analyzes and elaborates the vital technologies of target detection and tracking under single camera and explains algorithms' targeted parts.2.It respectively studies two kinds of handover algorithms under the two circumstances of overlapping horizons and non-overlapping horizons,which is aiming at the target handover technology in multiple cameras environment.In the case of overlapping horizons,it adopts the scale invariant feature description and homology matrix to verify the feasibility of automatic generation of horizon boundaries for target handover.In the case of non-overlapping horizons,it uses D-S(Dempster-Shafer)data fusion framework to fuse scale invariant feature,color histogram feature and UV chromaticity feature to complete target handover.3.In view of the wide range of surveillance and blind area of multiple cameras surveillance system,it studies to construct the topological relationship model between cameras.This model combines the camera monitoring system with Geographic Information System(GIS)to construct a camera network topology structure,and narrows the detection range of the target by time transfer probability between cameras,constructs the cost flow network of the target similar to the monitoring target in the camera monitoring area.In addition,a highest probability and the highest similarity target moving path is solved by the method of minimum cost and maximum flow.