Cross-Camera Vehicle Trajectory Estimation towards Traffic Flow

Cross-Camera vehicle tracking(CCVT) is an significant field in intellective transportation systems and intelligent visual perception in smart city. It is challenging due to the large variation of the same vehicle within various cameras. We propose a cross-camera multi-target vehicle tracking framework towards traffic flow parameter estimation, which is mainly divided into three steps:(i) multi-object tracking in a single camera, (ii) multi-camera trajectory association and, (iii) traffic flow parameter estimation. Among them, we design an encoder that integrates appearance and trajectory information for vehicle matching between multiple cameras. To demonstrate the validity of the algorithm, we collected multi-camera videos in real scenes of urban roads and annotated them manually. The experimental results demonstrate the effectiveness and accuracy of the proposed method.