Automatic Generation Algorithm And Implementation Of Cognitive Map For Autonomous Driving

In order to solve the high precision map based on lidar 3D point cloud,which is faced with the challenge of expensive on-board sensor and high computing power demand,the academic community proposed a cognitive map scheme with low cost,small data volume and moderate positioning accuracy.At present,cognitive map mainly relies on the geometric relationship between road signs and lane lines to complete vehicle positioning,and the local positioning accuracy needs to be further improved.In addition,the global positioning of vehicles faces challenges in the absence of road signs.Therefore,this paper first proposes a real-time vectorized map modeling scheme,which uses the vectorized map to replace the discrete point map to support the vehicle positioning scheme based on road geometry,thus improving the local positioning accuracy of vehicles.In order to solve the problem of vehicle global location,this paper adopts the location recognition algorithm based on the perspective of vehicle end,and realizes the location recognition scheme of vehicle global location estimation in large scale.The main research contents of this paper are as follows:(1)In order to support the local vehicle positioning scheme based on road continuity geometry,a real-time vectorized map modeling algorithm is proposed.The cognitive map is layered and divided into four layers: road layer,lane layer,semantic layer and dynamic information layer.The existing cognitive map lane layer uses discrete data points to represent the road,so the continuous geometric characteristics of the road cannot be obtained,which affects the accuracy of vehicle positioning.Therefore,this paper uses the vectorized map to replace the discrete point map in the lane layer,and proposes a real-time vectorized lane level map modeling algorithm to realize the automatic generation of the lane layer.(2)In order to solve the problem of vehicle global location,the location recognition scheme of vehicle global location estimation in large scale is realized based on the location recognition algorithm of vehicle perspective.The traditional automatic driving map needs to know the global position information of the vehicle at all times,otherwise it will not be able to realize the navigation function.This paper proposes that the location recognition algorithm based on the perspective of the vehicle can complete the global location estimation of the vehicle on a large scale only by relying on the vision,so as to serve the navigation function.(3)This paper uses the real-time vectored lane-level map modeling algorithm to complete the experiment of automatic lane layer generation in Qingshuihe Campus,and ensures its error within the centimeter-level range.This paper has completed the verification of the location recognition algorithm based on the perspective of vehicle end in Qingshuihe campus,and the recognition rate reaches more than 90%.