Lane detection and tracking using a new lane model and distance transform

Lane detection is a significant component of driver assistance systems. Highway-based lane departure warning solutions are in the market since the mid-1990s. However, improving and generalizing vision-based lane detection remains to be a challenging task until recently. Among various lane detection methods developed, strong lane models, based on the global assumption of lane shape, have shown robustness in detection results, but are lack of flexibility to various shapes of lane. On the contrary, weak lane models will be adaptable to different shapes, as well as to maintain robustness. Using a typical weak lane model, particle filtering of lane boundary points has been proved to be a robust way to localize lanes. Positions of boundary points are directly used as the tracked states in the current research. This paper introduces a new weak lane model with this particle filter-based approach. This new model parameterizes the relationship between points of left and right lane boundaries, and can be used to detect all types of lanes. Furthermore, a modified version of an Euclidean distance transform is applied on an edge map to provide information for boundary point detection. In comparison to an edge map, properties of this distance transform support improved lane detection, including a novel initialization and tracking method. This paper fully explains how the application of this distance transform greatly facilitates lane detection and tracking. Two lane tracking methods are also discussed while focusing on efficiency and robustness, respectively. Finally, the paper reports about experiments on lane detection and tracking, and comparisons with other methods.