Real-time Vehicle Status Perception Without Frame-based Segmentation for Smart Camera Network

Nowadays camera network plays an important role in the Intelligent Transportation System (ITS), and due to the weak computing ability of smart devices in the camera network, collecting traffic status in real time is one of the critical tasks in this field. A common strategy for traffic status collecting is first to form the trajectories of vehicles and then to measure interested indicators. To address this problem, we present a real-time vehicle status perception approach, which directly extracts vehicle status from our proposed novel video feature, temporal component-weight. Specifically, temporal component-weight is calculated based on a sampling of the whole frame. Also, a hybrid model is proposed to handle crowded situations. We test our approaches in surveillance sequences, and the results show that the proposed approach can effectively collect the vehicle status, including number, relative location, and relative speed.