Research On Ground Filtering And Road Recognition Method For Vehicle-borne Laser Point Cloud

With the rapid development of information age and "digital city"concept proposed, how to obtain 3D information quickly and accurately becomes an important problem to promote the construction of "digital city". The successful development of the vehicle-borne laser scanning system provides an effective means for solving this problem, which has been widely used in many areas such as digital city construction, traffic management and cadastral survey.Laser point cloud ground filtering is the first step in the subsequent classification of features. Compared with the ground filtering method of vehicle-borne data, the method of airborne data is more systematic and comprehensive, and how to apply the theoretical method for airborne data to vehicle data is an important development direction. In addition,the effective planning and management of transport facilities requires a large amount of geographic information data, so how to automatically identify roads from point cloud data containing multiple categories of objects will provide important help for the work of the transport sector.Based on the principle of airborne data ground filtering method, this paper has carried on the corresponding improvement on the analysis of the characteristics of vehicle-borne data, and realized the road recognition work on the basis of understanding the characteristics of urban road.The main contents and innovations are as follows:(1) Ground filtering method based on cloth simulation. This method is a new airborne laser point cloud ground filtering method proposed in recent years. Its idea is completely independent of the traditional ground filtering, and the effect is very good. In this paper, the method is implemented, and its feasibility and adaptability to vehicle-borne data are explored.(2) Improved progressive morphological laser point cloud ground filtering method. The traditional progressive morphological laser point cloud ground filtering Method is efficient and easy to be realized ,and the method in the airborne data has achieved good results. This paper analyzes the problems existing in the vehicle-borne data, and improves it from three aspects: ?using the adjacent ground point to fill the blank grid;?The terrain slope is estimated by constructing the triangulation network, and the progressive morphological morphological laser point cloud ground filtering is carried out one by one for each sub - region; ?According to the characteristics of vertical objects, different height difference threshold are set to extract the ground point.(3) A new road recognition method. In this paper, the boundary points of the road boundary is roughly extracted according to the morphological characteristics of the curb, and then the proximity clustering and the ordered least squares method are used to finely extract the boundary points of the road boundary. Finally, The pavement point cloud are identified according to road relationship.