Vehicle Localization Using Road Marking Image Matching (ICCAS 2018)

Accurate vehicle localization with map is an important task in urban environment. Currently, vehicle positioning with various types of maps has studied. In this paper, we propose a method which matches 2D-NDT map with road marking image for fast and accurate vehicle positioning. This method extracts features from occupancy grid map by Maximally Stable Extremal Region (MSER) detector. The extracted features on the map is modelled to centroid and covariance and these are stored on the map. This map contains little data but also key information. Camera images also require feature extraction as well as map to match with 2D-NDT map. Images of forward looking camera are converted to top view images via Inverse Perspective Mapping(IPM). Road markings can be accurately extracted by MSER in a bird’s eye view image. Vehicle Localization is performed by NDT map matching between the map and road markings in a bird’s eye view road image. The evaluation of this method was conducted on a straight road in urban environment. Vehicle positioning results revealed accurate lateral, longitudinal position estimation and orientation estimation. Since NDT map matching was done with only significant lanes on the road, matching speed was fast.