Robust multi-lane detection and tracking for automated highway driving system

Abstract

Since the early 2000s, many global automotive industries have been working to reduce traffic accidents and improve driver convenience by developing autonomous vehicles. Lane detection is especially necessary for the most basic and important advanced driver assistance system (ADAS) such as lane keeping assist system, lane change assist system and lane departure warning system. However, although existing algorithms have good performance in lane detection on general roads, when there are poor road markings or surrounding obstacles exist on the road, a false lane detection occurs or lane detection is not performed properly. Therefore, in this paper, we propose a robust multi-lane detection and tracking algorithm for automated highway driving. To solve the above problems, we introduce three core technologies. First, using a local adaptive threshold, we find features of lanes that are difficult to extract and remove noises caused by obstacles. Next, we use an adaptive RANSAC algorithm to extract only a correct lane. Through the adaptive RANSAC algorithm, the correct lane is detected except for outlier by using a location and angles of edges. Also, a curvature of lane history is used to prevent a false lane. Finally, through the lane classification, the accuracy of the detected lanes is greatly improved by detecting only lanes that are verified once again. This algorithm achieved 99.9% precision and 99.0% recall using a general highway dataset.