Research And Implemention For The Method Of Road Boundary Detection And Road Marking Rescognization Based On Point Cloud Of Lidar

With the development of autonomous vehicles and people's increasing concern about traffic safety.Road boundary detection and road marking recognition play significant roles in the safe driving of unmanned vehicles.If unmanned vehicles accurately can detect the road boundary,unmanned vehicles can distinguish the driving area in traffic environment,and if unmanned vehicles can recognize the type of road marking,unmanned vehicles can determine the current lane information,and thus unmanned vehicles can accurately make sure the behavior of going straight,changing lanes or turning and doing other driving behavior.There is a mainstream research that contains road boundary detection and road marking recognition,one is detection based on visual inspection,the another is detection based on the LiDAR.There are cameras vulnerable to the influence of external light environment based on visual inspection,especially the shadows and water exist in the environment,and light conditions change,these situations can cause the image is in low clarity and blur distortion,detection results are easy to appear deviation and even cannot be detected.However,LiDAR can obtain high precision,high density information of point cloud with strong penetrating power,and it is not affected by light conditions change,so LiDAR has a larger advantage in shape and brightness detection.At present,in the research of detecting road boundary and recognizing road markings based on the point cloud,many methods use some characteristics,such as local characteristics of the road boundary and high reflection characteristics of road markings,the method is in high real-time performance,good robustness and it can obtain good detection results and recognition results.There are two parts in the research.The first part is for road boundary detection,the elevation and slope change are bigger in the general road boundary area,but there are green belts in some road boundary area,leading to a cluttered change to the geometrical characteristics of the road boundary,the proposed algorithm of road boundary detection get accurate results for the road boundary blocked or without shade,the main steps are as follows,road boundary detection algorithm firstly obtains road cross sections,and then it adopts the method that using a single-line to get the road boundary points in road cross sections from bottom to up,in this way it can get cross sections contours,then the algorithm can determine road boundary points through the double-window method,which can detect the elevation and slope characteristics of road boundary,finally the algorithm obtains road boundary lines accurately by fitting the road boundary points.The algorithm introduces a single-line point cloud to extract road contours.Even if when the road boundary is obscured,the algorithm can also protect road boundary shape characteristics,which can enhances the robustness of the algorithm.The second part is recognition of road markings,first segmenting the road area according to road boundary lines,and then using dynamic-threshold algorithm to extract the road markings based on the intensity characteristics,and then clustering various road markings through hierarchical clustering algorithm,after that computing the length and width of clusters by using PCA algorithm,and determining planeness of clusters by kd-tree,the classification of road markings would be determined by these three characteristics,finally determining the center of the road markings and nested the model of road marking.As is shown by experimental results,the method can accurately detect the road boundary,and recognize the types of road markings in the urban traffic environment,and has a good practicability.