Research On Vehicle And Pedestrian Detection And Ranging Method Using Vision Fusion

Advanced Driver Assistance Systems(ADAS)is one of the most important research directions in the field of Intelligent and Connected Vehicles(ICV).Vehicle and pedestrian detection technology is also an important part of ADAS.Aiming at the existing vision-based vehicle and pedestrian detection methods,there are some problems such as low accuracy,poor real-time performance and even difficult to identify the target.This paper researchs on vehicle and pedestrian detection and ranging method using vision fusion.Firstly,this paper introduces the research of vision-based vehicle and pedestrian detection methods at home and abroad from the aspects of vehicle and pedestrian appearance and motion information.According to the theory of image processing,the reasons of low detection rate and poor real time performance of common vision-based vehicle and pedestrian detection methods are analyzed.This paper researchs on vehicle and pedestrian detection and ranging method using vision fusion.Then,an adaptive ROI segmentation model and its algorithm for vehicle and pedestrian detection based on road horizon are presented.KITTI(Karlsruhe Institute of Technology and Toyota technological)datasets included vehicles and pedestrians are targeted.In Visual Studio 2013 software Environment,the ROI is segmented,and the ROI image of vehicle and pedestrian is obtained.After,according to the theory of deep learning object detection based on multilevel features and the theory of object detection using U-V disparity map based on object plane characteristic.Vehicle and pedestrian detector and range finder based on SSD(Single Shot Multibox Detector)deep learning model and U-V disparity map are designed.The vehicle and pedestrian detection and ranging experiment of ROI image is carried out by using the detector and rangefinder.Vehicle and pedestrian object detection and ranging method using vision fusion compared with common visual-based vehicle and pedestrian object detection methods,the experimental results show that the detection speed of this method increased by 24.1%,the detection rate was increased by 1.46%,the missing rate was increased by 2.83%,the false detection rate wase reduced by 1.10%,the ranging error of vehicle and pedestrian within 30 m is less than 10.5%.