| Positioning and navigation is a key part of the driverless system and an important topic in the research of autonomous driving technology.At present,GPS-based navigation methods are widely used in the field of autonomous driving,and have achieved great success.However,the operating environment of autonomous vehicles is complicated and diverse,and not all have good GPS signals.Especially in the GPS signal rejection environment,autonomous vehicles that rely solely on GPS will face the danger of navigation failure.At this time,if the high-precision map of the scene is known,the high-precision positioning of the autonomous vehicle by means of a sensor such as a laser radar can also complete the navigation task.However,for a previously unknown application environment,the construction and maintenance of high-precision maps and the purchase of laser radars require a large cost,which greatly limits the application range of high-precision positioning navigation.The ultimate goal of the studying of driverless is to replace humans in performing tasks in various environments.However,the reality is that driverless cars still have a long way to go to truly replace humans.As far as navigation is concerned,the previously described "harsh" navigation conditions for autonomous vehicles are not so difficult for humans.This paper provides an in-depth analysis of this and proposes a humanoid navigation method for autonomous vehicles in GPS signal denied environment.And the feasibility and effectiveness of the proposed algorithm are verified by design simulation and real vehicle experiments.The main work completed in this paper is as follows:1.We propose an autonomous vehicle positioning method based on "inflection point" matching.Firstly,according to road network data of the scene and the annotation of candidate inflection points,we analysis the relationship between inflection points and the surrounding road points,extract describe vector of the inflection points,build the inflection points database.Secondly,we record real-time trajectory of the autonomous vehicles using visual odometry,and analyze the feature representation of the current trajectory point in the same way.Then we match its representation vector with the feature vectors in the database.According to the matching results and the position of the former inflection point to estimate which road the autonomous vehicles is on in the road network of the scene.And we also reset the visual odometry at the inflection point to eliminate its cumulative error.2.We propose a method to plan the moving direction of the autonomous vehicles based on multi-cue fusion.Firstly,a variety of cues related to the moving direction of the autonomous vehicle are extracted from the scene image,and then gaussian model about the moving direction of the autonomous vehicle is constructed respectively.Finally,the planning results of different cues are integrated with the Bayesian framework to guide the next steering of the autonomous vehicle.3.Integrated navigation experiment of the autonomous vehicle under GPS signal denied condition is designed in our campus scenes,including the simulation and real vehicle data experiment of the autonomous vehicle positioning based on the "inflection point" matching method,as well as the simulation and real vehicle data experiment of the moving direction planning of autonomous vehicle based on multi-cue fusion.The feasibility and effectiveness of the humanoid navigation method in this paper are verified through the experiments. |
