| With the need for transportation in today’s society,traffic has increased faster then roads can be built.This has cause problems:safety on the road and traffic congestion are the main culprits in stopping the efficiency of social production activities.Self-driving vehicles could be used to help the problem as they are made to help with these issues.They have a better awareness of what is happening round them and are also able to plan where they are going automatically.These improvements can help reduce accidents and traffic jams in the major areas of cities and towns.The biggest difficulty to this scheme is the making of the environmental recognition software,which is the whole point of the idea.In this paper,the research into the ability of making such a program is studied,as well as the environmental awareness framework for autonomous vehicles through different sensors recognising low and high level road environments.The information on this study are as follows:First of all,LiDAR sensors are used for the recognition of the low-level road environment.First,analysis the properties of pointcloud at edge of road by analogy with two dimensional Sobel algorithm.Then a filtering algorithm with the analysis results filter the pointcloud twice.The pointclouds at the road edge are then selected.The false positive point are gotten rid of.by combining the travel path of the autonomous vehicle,together with the shapes of the line on the road side and the RANSAC idea.This will make the roadside line mappable and divide the driverless areas into driving areas.Secondly,high-level road recognition is the next step.By introducing deep learning and camera sensor to the semantic segmentation of the lay out of the road.By studying and analysis of the urban road environment,a deep road scene segmentation network made of of 13 layers convolutional encoded-network and 13 layers deconvolution decoded-network based on the convolution neural network is shown.In effective conclusion to this.In order to make the model effective semantic segmentation of road scene images in the campus environment,we adjust the model by the way of transfer learning.The road speed image can now effectively can be semantically segmented,even road scene content can be classified exactly for recognition at pixilated level.Finally,the accuracy of the way for environment has been tested under the conditions of the campus area.The low and high level road environment recognition software has been achieved,and the frame of perception of a driverless environment has been built. |
