Konferens: FAST-zero'15: 3rd International symposium on future active safety technology toward zero traffic accidents, 2015, Gothenburg

We tried to develop the new localization method by simple 2D-plane map without deterioration of estimation accuracy. The boundary line has a lot of features e.g. changes of height, color and brightness, but they are sensitive for noises. From the robustness point of view, it is difficult to match the road boundary line with the boundary line on 2D map. The localization method using 3D point cloud matching or texture matching are so accurate, but these have disadvantage in adaptation to the change of environment. So, we decide to make the classifier to classify as road area or the other area, and propose the new localization method that has advantage in robustness by matching the identified shape of road area with the shape of the road on 2D plane map. First, we calculate the HOG features from the range data acquired by 3D LiDAR. Then, we make the road plane classifier applying SVM.