Research And Simulation Of Driverless Perceptual Aids System

With the rapid development of computer hardware and software,driverless has become a cutting-edge technology that has been stepping up research and development at home and abroad.Most automakers today promise to be semi-automatic by 2020,and many experts in related fields predict that fully automated vehicles will become possible in the near future.Mature driverless technology will not only improve the traffic efficiency of the whole society,but also reduce the occurrence of traffic accidents and save more lives.As a basic component of the driverless system,the perception system directly affects the effects of the driverless decision system and the control execution system.The perceptual system mainly uses computer vision technology to extract the information in the operating environment of the unmanned vehicle,including the detection of the lane line,the pedestrians in front of the unmanned vehicle,the identification and tracking of the vehicle,etc.And can be used to assist unmanned driving,support for decision making and control execution systems based on sensing capabilities.Based on the realization of the relevant perceptual algorithm,this paper proposes a series of early warning standards based on the actual situation of unmanned driving,and completes the driverless perceptual assistance system.At the same time,the application field of virtual simulation technology is more and more extensive.In view of the high cost and uncontrollable factors for testing the effect of the cognitive aid system on the real road,this paper builds an unmanned virtual simulation platform based on Web and visual simulation platform Unreal Engine.(1)Research and implementation of lane line detection algorithm.In view of the particularity of lane line in color,texture and position,this paper combines these features to perform feature point pre-extraction according to multiple adaptive threshold criteria.Because traditional method uses hough transform,approximate the lane line with a straight line approximation is not applicable to the scene with a large curve.In this paper,the local dynamic window is used to select the quantitative and evenly distributed fitting points,and the lane line is fitted by polynomial.(2)Research and implement forward vehicle and pedestrian detection algorithms.The visual word bag is used as the basis for classifier training and testing.In view of the background information will interfere with the target detection,this paper proposes to use the OTSU-based image segmentation method to extract the image target region and extract the word bag feature from the target region.In order to ensure the uniform distribution and independence of visual vocabulary,avoid confuse visual vocabulary and affect the detection results,this paper proposes a visual dictionary construction method based on category information.The experimental results show that extracting visual vocabulary based on categories to target areas can improve detection accuracy.(3)In view of the uncontrollability of testing the perceptual aid system on the real road,this paper designs and implements an unmanned virtual simulation platform based on the combination of Web and Unreal Engine.Through the different choices of users in the browser,different virtual simulation environments can be dynamically created on the UE side.The virtual simulation platform combines the friendly human-computer interaction features of the Web with the powerful visual simulation features of Unreal Engine.(4)The unmanned perceptual aid system realized in this paper is applied to the virtual simulation platform for simulation experiments.Different environmental variables are set in the experiment.After repeated testing and analysis,the experimental results show that the proposed assistant system is feasible and effective.