Mapping And Localization Method Based On Visual-inertial SLAM For Intelligent Vehicles

Localization is one of the most important techniques in the research of intelligent vehicle.The most common localization method is to use high-precision GPS,while it cannot be used in GPS-denied areas.Simultaneous localization and mapping(SLAM)on intelligent vehicles can deal with this problem.Since cameras are low-cost,meanwhile can provide rich information,visual SLAM has good prospects.This thesis focuses on visual mapping and localization for intelligent vehicle.The current vSLAM methods have several limits,the first is that the motion estimation of vehicles is not accurate enough since most methods are not especially for vehicles;the second is that they can not properly deal with large-scale mapping;the third is the lack of complete system for mapping and localization.To solve these problems,this thesis proposes a visual mapping and localization method based on keyframe viSLAM for intelligent vehicle.The contributions can be divided into three parts.Firstly,a odometer-aided viSLAM method with feature selection in adaptive grids is presented to improve the accuracy of vehicle motion estimation.Secondly,a hybrid mapping method with viSLAM and high-precision GPS is proposed to uniformly build visual feature maps in GPS-denied areas and GPS-available areas.Lastly,this thesis use an optimization based fusion localization method to utilize the information of localization with feature matching and local motion estimation from viSLAM.The experiments show the state-of-the-art performance on local vehicle motion estimation,mapping and localization of the proposed method.