Convolutional Network and Moving Object Analysis for Vehicle Detection in Highway Surveillance Videos

Moving object detection is an underlying task in so many applications, especially video surveillance applications. One of the applications is for detecting moving vehicles on the highway. The detection of moving vehicles on the highway can then be further analyzed for intelligent transportation systems, such as vehicle counting, tracking, and classification. In the proposed method, the improved background subtraction through bilateral filter, morphology, and connected component is employed to detect moving vehicles preliminarily. Then, the obtained region of moving vehicles is fed into convolutional neural network-based YOLOv3 to detect the location and type of the vehicle. Two highway surveillance videos, including CDNET 2014-“highway” and a public highway video (from YouTube) are used to verify the proposed method. Our approach obtains significant improvement in terms of quantitative and qualitative results. Specifically, the obtained f. measure and percentage of wrong classification (PWC) are 0.8831 and 3.5234, respectively.