Research On High-definition Map Generation Method For Autonomous Driving

Autonomous driving is an important tendency in the area of artificial intelligence technology,and also one of the new strategic directions of China.Nowadays,autonomous driving has become a hot topic for academia and industry.High-definition map can effectively improve the safety and stability of intelligent vehicles,and it is one of the key factors for future intelligent travel.However,the traditional navigation electronic map and high-definition map are quite different in data model,geometric expression,accuracy,scale,supporting for dynamic expression,etc.The existing high-definition map models have some defects,such as incomplete data model,inconsistent map format expression and poor scalability,which severely restrict the popularization and development of autonomous driving technology.This paper studies the high-definition map model and automatic generation methods of autonomous driving,which provides the key supporting technology for autonomous driving.To be specific,the main contents of the paper are as follows:(1)Firstly,the background and significance of the paper are introduced,and the research contents of the paper are presented.The paper summarizes the generation and development of the high-definition map from three aspects,the existing navigation electronic maps,the development situation of autonomous driving abroad and the application research of the highdefinition map in autonomous driving.Traditional navigation electronic maps are not suitable for autonomous driving,and data representations of the existing high-definition map models are inconsistent.According to these problems,the research objectives and approach of the paper are put forward.(2)Traditional navigation electronic maps are not suitable for autonomous driving,and data representations of the existing high-definition map models are inconsistent.To solve this problem,a high-definition map model for autonomous driving(intelligent definition map model)is proposed.Firstly,the problems of the existing high-definition map models are analyzed.Then,the concept and structure of the intelligent definition map model are described in detail.The model puts emphasis on the integrity of data attribute and structure,the definition of data content,dynamic expression of spatial geographic information,the extensibility of information.Finally,the connotation of road network layer and generalized POI layer of the intelligent definition map model are elaborated.(3)For the lane-level road network automatic generation of the high-definition map,a fast lane-level road network generation method based on trajectory similarity-join pruning strategy is proposed.Firstly,calculate the nearest distance point on the segment centerline trajectory for each lane centerline trajectory point.Then,calculate the shape point of each lane centerline according to the formula of fixed score point.For calculating the nearest point and the nearest distance from the trajectory to trajectory and the corresponding symmetrical points,a fast pruning method is proposed.To calculate the nearest point and the nearest distance of the lane trajectory points,twice pruning are conducted by using the trajectory for the parallel relationship between the trajectory-similarity-join.This provides a new map data source and a new approach of map generation for map manufacturers.(4)The method of automatic construction of lane-level road network topology based on multi-directional constraints PCA is proposed for the high-definition map.Firstly,PCA projection is carried out based on the clustered point sets to obtain the direction and set of the road network layer.Then,based on the direction constraint,the physically continuous lane network is separated and the linear event points are extracted.Based on linear events,the physically continuous lane network is segmented and the correlation between lane and road network is constructed.Finally,the topological relationship of intersections is established.(5)The intelligent definition map model,the automatic generation method of highdefinition map and the automatic topology generation method of high-definition map are experimented and validated by using the map data of the real world.Different experimental methods are used to analyze the results of lane-level road network generation in different scales.The results show that the approach proposed by the paper can make full use of the existing professional surveying and mapping data of road-level road network and autonomous driving vehicle data to quickly generate lane-level road network.Through qualitative attribute analysis and quantitative accuracy evaluation of the experimental results,the proposed method can generate high-definition lane-level road networks with correct and reliable topological attributes.