Research On Information Perception And Driving Behavior Planning Of Unmanned Intelligent Vehicles At Night

The unmanned intelligent vehicle system is a comprehensive intelligent system integrating environmental sensing,adaptive path planning and driving behavior planning.With the rapid development of industrial information technology,the unmanned intelligent vehicle system has been continuously innovated and gradually entered the public’s field of vision.At the same time,the driving safety of unmanned intelligent vehicles has also received widespread attention.Unmanned intelligent vehicles can quickly and accurately perceive vehicles,pedestrians and other objects in the surrounding environment which is an important prerequisite for safe driving on the roads at night.To achieve fast and accurate environmental perception,the primary problem is to provide unmanned intelligent vehicles with the right sensors.Because the different sensors have different advantages and disadvantages,researching a three-dimensional target detection algorithm based on multi-sensor fusion,and giving full play to the characteristics of each sensor are of great significance for improving the information perception ability of unmanned intelligent vehicles.At night,another important part of the safe driving of unmanned smart cars on the road is the realization of driving behavior planning ability.When the unmanned intelligent car senses the information of the surrounding complex environment,it needs to filter and judge the complex information and take corresponding strategies,so as to drive autonomously,it is an important issue for the driverless intelligent vehicle to make reasonable and driving behavior planning independently.In order to improve the information perception ability and driving behavior planning ability of the unmanned intelligent car in the night environment,this paper deeply studies algorithms of threedimensional target detection and driving behavior planning of the unmanned intelligent car.Therefore,the research content of this thesis mainly includes two parts: one is the multi-view channel three-dimensional target detection algorithm based on infrared image and lidar point cloud fusion;the other is the driving behavior planning algorithm based on cone point cloud and reverse reinforcement learning.The main innovations of the paper are as follows:1.In this paper,a three-dimensional target detection algorithm based on infrared image and lidar point cloud fusion is proposed.Aiming at the inefficient detection of surrounding objects by unmanned vehicles at night with infrared image,the algorithm combines the infrared image data and the synchronized Lidar point cloud data,and the advantages of complementary information of two different sensors are utilized to greatly improve the accuracy of target detection.The multi-view channel fusion network is improved,and a method of coding and decoding based on feature pyramid is proposed.The method can learn to remap the feature map back to the original input size,so that the feature map output by the feature extraction network has high resolution and high representation,and it can improve the detection effect of the small target,especially the pedestrian.2.This paper proposes an unmanned vehicle driving behavior planning algorithm based on hierarchical reverse reinforcement learning.In view of the problem of autonomous driving behavior planning of unmanned vehicles under night conditions,the stratification strategy adopted by humans in carrying out driving behaviors is considered.The planning process is divided into two levels: discrete and continuous.The reverse reinforcement learning method is used to learn the real driving behavior of human beings,and the conditional distribution of all possible future driving behaviors is expressed as a mixture of probability distributions.The final experimental results show that the algorithm can fit the driving behavior of a human driver very well.