LaneMapper: A City-scale Lane Map Generator for Autonomous Driving

Abstract

Autonomous vehicles require lane maps to help navigate from a start to a goal position in a safe, comfortable and quick manner. A lane map represents a set of features inherent to the road, such as lanes, stop signs, traffic lights, and intersections. We present a novel approach to detect multiple lane boundaries and traffic signs to create a 3D city-scale map of the driving environment. We detect, recognize and track lane boundaries with multimodal sensory and prior inputs, such as camera, LiDAR, and GPS/IMU, to assist autonomous driving. We detect and classify traffic signs from the image considering high reflectivity of LiDAR points and further register the locations of traffic signs and lane boundaries together in the world coordinate frame. We have also made our code base open-source for the research community to tweak or use our algorithm for their purposes.