Mapping Lane Markings with Multi-Sensor Data

This paper presents a map generation process based on a low-level camera lidar fusion algorithm. The system uses images from RGB cameras and lidar pointclouds to calculate the exact position of the lanes, which are combined with GPS data to generate an accurate map in a global coordinate system. The lanes are identified by lane markings and are also assigned various features such as; the lane width, radius of the curvature and line type. It also presents how the algorithm fuses the camera image and the pointcloud and then generates the spatial points of the bands. It shows how the system generates the descriptors associated with the lanes and the markings. It points out the source of the noise that can be observed in the system, and the possibilities for future improvements.