We present a novel framework formultiple pedestrian tracking using overlapping cameras inwhich the problems of object detection\nand data association are solved alternately. In each round of our algorithm, the people are detected by inference on a factor\ngraph model at each time slice. The outputs of the inference, namely, the probabilistic occupancy maps, are used to define a cost\nnetwork model. Data association is achieved by solving a min-cost flow problem on the resulting network model. The outputs\nof the data association, namely, the ground occupancy maps, are used to control the size of factors in graph model in the next\nround. By alternating between object detection and data association, a desirable compromise between complexity and accuracy is\nobtained. Experiments results on public datasets demonstrate the competitiveness of our method compared with other state-ofthe-\nart approaches.
Loading....