A novel video analysis approach for overtaking vehicle detection

Road safety plays a pivotal role in intelligent transportation systems. In the past decade, much attention from researchers and industries alike has been directed towards intelligent systems that can help reduce the number of on-road accidents. For obstacle detection applications, sensor-based approaches employing radar and lidar are appealing as they offer high object detection rates. However, videos are a much richer source of information and video-based approaches can offer a holistic view of the scene, bringing about new applications that are not limited to just obstacle detection. In this paper, a novel video-based framework is proposed for detecting possibly overtaking vehicles, which may ultimately lead to a potential hazard. The system relies on lane marks to guide the vehicle tracking module and leverages the information extracted from the vanishing point to correctly distinguish a vehicle passing by from the one truly attempting to overtake the ego-vehicle. The key to the success of the system lies in the accurate detection and tracking of the trajectory of the vehicle, localization of the lane marks, and estimation of the vanishing point. Experimental results obtained from running our algorithm on real-world footage confirm that the proposed system produces high detection and low false alarm rates.