Visual Guidance Algorithm Of Unmanned Vehicle Without Manual Labeling

Mainstream vehicle autonomous driving technology mostly rely on high-precision positioning or accurate road perception.However,due to the influence of environment,weather,illumination and other factors,it is still impossible to obtain accurate positioning information and accurate road perception information for all driving scenarios.In view of this,this paper based on independent high-precision map and high-precision positioning,carries out research on the visual guidance algorithm of unmanned vehicles in the face of changeable road scenes without manual labeling,and realizes autonomous driving of unmanned vehicles based on the visual guidance algorithm.The main work and achievements of this paper are as follows:1)Aiming at the dependence of mainstream autonomous driving schemes on highprecision positioning and accurate road perception and the lack of interpretation and unpredictability of end-to-end autonomous driving algorithms,this paper proposes a new idea of vehicle autonomous driving based on path guidance points.Under this framework,this paper proposes two automatic labeling methods of path guidance point labels based on visual odometer and wheel odometer/IMU combination,achieving nearly "zero cost" of labeling data acquisition,providing strong data support for the subsequent deep learning-based path guidance point generation algorithm.2)Aiming at the combination of environmental perception information and decision planning information,this paper designs a path guidance point generation algorithm which integrates local navigation map and front view image.The algorithm realizes the crossmodal information fusion of the image from the bird’s-eye view and the visual image from the forward view,effectively combines the environmental perception information with the location information,and makes the path guidance point generation algorithm guided by the global planning information.The proposed algorithm is tested on the data set of CARLA simulation environment and KITTI data set based on real vehicle environment.The experimental results confirm the effectiveness of the algorithm.At the same time,compared with the traditional fusion method,the fusion strategy designed in this paper brings 5.7%and 1.5%accuracy index improvement respectively in the above two kinds of data sets.3)Based on the above research results,this paper completed a set of unmanned vehicle autonomous driving system based on route guidance point demonstration prototype development,and in the campus environment of the demonstration prototype has carried on the real vehicle tests,proved in this paper,based on route guidance point of autonomous driving system can quickly realize unmanned vehicle in the form of low-cost autonomous driving,this paper highlights the practical.