Multiple Target Detection and Tracking in a Multiple Camera Network

Abstract: Given synchronized video sequences from a number of cameras withoverlapping fields of view, the detection and tracking of a prioriunknownnumber of individuals entering a determined area is considered,showingthat a generative model can accurately follow the individuals and handleeffectively such problems as occlusions in each view independently. Theaim of this thesis is to implement the exchange of information betweenthe cameras where the detection and tracking processes take place. Theinputs are obtained from synchronized videos and the frames are takenindividually to treat them as independent images. The proposed algo-rithm was implemented in MATLAB and results obtained on a personalcomputer are presented. The results show that the algorithm achievesgood tracking accuracy, has relatively low computational complexity, andat the same time it allows to observe the communication requirementsbetween the cameras and the processing node.