Research On Anti-collision In Night Vision System Of Autonomous Vehicles

In recent years,with the rapid development of computer hardware and the crazy growth of data volume,the society has entered the era of big data,artificial intelligence is sweeping all kinds of industries,driverless cars are thedirection of automobile industry.China's new generation of artificial intelligence development plans and major science and technology projects was launched in Beijing on November 15,2017,in which Baidu has become one of the first national artificial intelligence open innovation platforms based on its driverless technology.This fully shows that the nation attaches great importance to artificial intelligence technology such as unmanned driving.Anti-collision technology is an important part of unmanned research,and because of the lack of light at night,it is a regular time for collision events.Therefore,the technology of collision prevention in night vision environment is the most momentous.Based on the practical problems of night driving such as the lack of light,this paper studied the collision prevention technology of night vision environment for driverless vehicles.Firstly,this paper denoised and enhanced the pictures transmitted by the night vision device in the car,then detects the ROI region of the suspected obstacles on the night vision picture.Secondly,this paper used trained classification model classifying obstacles ROI image,then designed a accurate real-time stability can make reasonable and humanized collision decision which is based on different obstacles categories and the safe distance.Finally,the theoretical and technical research in this paper was carried out in two stages of PC+Matlab environment and laboratory car RoboCAR simulation and verification.The specific research contents are as follows:?1?Research on algorithms of preprocessing in night vision image:In order to solve the common problems in night vision images such as poor equality of image,the algorithms of noise reduction and image enhancement in night vision image were proposed,and the ROI region of suspected obstacles was detected.?2?Research on models of obstacle classification:In order to make a better anti-collision strategy,two kinds of obstacle classification models based on HOG+SVM and deep learning were designed,which can distinguish the main obstacle from the secondary obstacle,and the performances of the two models were tested.?3?Research on anti-collision decision-making based on the safety distance:This part includes the method of the distance between the vehicle and obstacle,the measurement of vehicle speed and a model based on safety distance in which the first-level judgment was made by the presence of obstacles called O while the secondary judgment was made by the categories of obstacle called C and the safety distance calleds₀.There are specificly four kinds of intelligent decision-making countermeasures to be realized:no intervention,flashing lights and whistle,speed deceleration tov₀,parking.?4?The experimental results of simulation and verification:The theoretical techniques involved in this project were simulated and verified in two phases in the PC+Matlab environment and the laboratory car RoboCAR.The experimental results showed the feasibility of the algorithm and the accuracy of the model.