Understanding surrounding vehicles in urban traffic scenarios based on a low-cost lane graph

Understanding the surrounding vehicles with respect to the road lane context and the host vehicle is crucial for the success of autonomous driving and advanced driver assistance systems (ADAS) used in daily urban traffic. This paper proposed a vision-based approach to locate the surrounding vehicles into the corresponding driving lanes as well as classify them with respect to the host vehicle´s driving. In particular, the driving corridor of the host vehicle is generated by fusing multiple sources of data and the leader vehicle is determined, if available, to provide the real-time and validated information, such as trajectory and velocity, for the control purposes of the host vehicle. The potential applications of such information range from extending the functionality of the existing ADAS systems, e.g., vision-based platooning, stop-and-go traffic jam assist, etc., to autonomous driving in terms of high-level decision making and path planning. Furthermore, instead of the expensive high-precision detailed map, only a simple lane graph, constructed automatically by using conventional low-cost sensors, is used in this work to provide the driving lane information. Experimental results in various typical but challenging urban traffic scenes have substantiated the effectiveness of the proposed system.