In this paper, we propose a novel distance-based camera network topologyinference method for efficient person re-identification. To this end, we firstcalibrate each camera and estimate relative scales between cameras. Using thecalibration results of multiple cameras, we calculate the speed of each personand infer the distance between cameras to generate distance-based cameranetwork topology. The proposed distance-based topology can be appliedadaptively to each person according to its speed and handle diverse transitiontime of people between non-overlapping cameras. To validate the proposedmethod, we tested the proposed method using an open person re-identificationdataset and compared to state-of-the-art methods. The experimental results showthat the proposed method is effective for person re-identification in thelarge-scale camera network with various people transition time.
展开▼
