Vehicle High-Precision Positioning Technology Based On Visual Perception Under GPS Blind Area

Vehicle high-precision positioning is a fundamental and critical technology to realize intelligent vehicle.The common GPS has limited accuracy,location data cannot be outputted in satellite obscured situation,such as urban roads lined with tall buildings,underground tunnels,viaduct bottom and so on,and it is unable to meet the positioning demand of all-weather and no area restriction.Therefore,through analyzing in-depth vehicle high-precision positioning technology,on the basis of not relying on GPS device,visual perception algorithm is used to estimate vehicle location in paper.The study establishes a vehicle positioning system based on multi-mode synergy,and some key technologies,such as image matching with illumination robustness,vehicle relative positioning based on visual odometer and vehicle absolute positioning based on visual map building,are researched in-depth.Main contributions of the paper include the following four aspects:(1)The image matching algorithm with illumination robustness is put forward.Aiming at changeable pavement images in driving environment,a single matching algorithm cannot be used to obtain accurately feature,so the paper uses respectively Harris,SUSAN,FAST,SIFT and SURF to process a variety of road images under normal illumination conditions.Their actual application are decided according to detection rate and runtime.Due to the interference of illumination changes existing in the collected images,many traditional matching algorithms under illumination change are not optimal,so the paper proposes an image matching algorithm with illumination robustness.We know image edges and detail information have lower sensitivity for illumination change,SURF feature points are optimized by image gradient based on the idea of Canny,and bidirectional search is used to obtain matching points.The experimental results show that feature point detection of the algorithm still remains good stability for brightness change.Compared with the traditional algorithms above,it is robust for illumination change,ensures the higher matching speed,and meanwhile improves greatly the matching accuracy.(2)Vehicle positioning algorithm based on the fusion of road feature matching and optical flow is presented.Firstly,in allusion to two situations of larger offset and less offset between road images,vehicle positioning algorithms based on feature matching and improved Lucas-Kanade are respectively suggested.Aiming at different mismatching results,PROSAC and the custom LARSAE are used to optimize.On the basis of analyzing their matching accuracy,frequency and running time,the discrete kalman filter is applied to fuse two algorithms,and vehicle motion error estimation is used to correct location,which overcomes the low accuracy of optical flow and longer processing time of feature matching.Through comparison and analysis of several experiments,two algorithms based on feature matching and optical flow have higher accuracy and more smooth than GPS positioning,and it can effectively solve the problem of vehicle positioning under GPS blind area,to ensure the continuity of vehicle trajectory,but it will produce the accumulative error over time.Meanwhile,Fusion algorithm has higher positioning accuracy as well as considering real-time,has better effect in the flat road,and can provide a more precise result with stronger noise resistance under illumination change and unclear ground texture.(3)Vehicle positioning algorithm based on visual map building is proposed.In order to radically eliminate cumulative error of relative positioning,the paper brings in the static environment feature to combine vehicle localization and map building.Namely,vehicle sensors are used to create map,and then it is applied to achieve accurate positioning.Firstly,visual database is built,including scene images,dominant color,SURF feature and their corresponding locations.Then the on-board camera shoots vertically a scene image in front,several similar images can be obtained based on the dominant color feature and DP matching,and weighted fusion is executed to realize accurate positioning by combining with their eight neighborhoods.The experimental results show that the time of constructing database is too much,but data amount is reduced in the process of vehicle positioning,and its precision meets the actual needs.Meanwhile,the algorithm does not drift with time,and is suitable for vehicle random positioning in a fixed line.The positioning system has low cost,high reliability,strong resistance to interference and practical value.(4)This paper constructs a vehicle positioning system based on multi-mode synergy.Through analyzing the requirements of vehicle positioning system,the software and hardware platforms are set up.Visual sensors are used to collect road and scene image,vehicle positioning technologies based on visual odometry and visual map building are cooperated to estimate vehicle movement,the error correction device based on vehicle detecting and RFID perception is designed to optimize positioning results,and finally the optimal correlation of multi-group location information is realized.The system verifies the reliability and applicability of the co-location algorithm based on intelligence vehicle.The experimental results show that the positioning system can compensate for the shortage of GPS positioning,and has a crucial connection effect for obtaining vehicle accurate location information of all-weather and no region restriction.Vehicle positioning algorithm analysis and vehicle test show that the positioning accuracy of the proposed algorithm can reach up to 0.5m,and it makes up the limitation of GPS sensor,to solve vehicle positioning problem in the special environment of the GPS blind area.In a word,the study provides the basis of theoretical research for the practical application of intelligent vehicle system in the future.