Research On Indoor Location System Based On RGB-D Camera

Simultaneous localization and mapping(SLAM)is the key technology of autonomous motion on robot,which has a wide application prospect.With the rapid development of robot technology,it's very important that how to improve the accuracy of robot location.Because the RGB-D camera has a comprehensive ability to obtain information,the SLAM system based on RGB-D camera has become a research hotspot of vision SLAM technology.In this dissertation,the point feature and line feature are extracted from RGBD images and will be used to construct the plane feature.The point feature,line feature and plane feature are fully fused.A vision SLAM method based on the point feature,line feature and plane feature is proposed,which can be used for robot positioning and attitude estimation in the structured indoor environment.First,this dissertation improves the process of point feature extraction to solve the problem of point feature block.In the extraction part,an AT-FAST algorithm based on adaptive threshold is designed to solve the point feature block,and the point feature matching is carried out by RANSAC algorithm.Secondly,we improve the accuracy of line feature extraction to solve the problem that high redundancy of line feature leads to low matching efficiency.The extracted line features are optimized,the short line segments are removed and the split line segments are merged.The LBD descriptor is used to describe the line features in the feature matching part,and the binary representation is used to reduce the computational complexity of matching.Then the candidate plane is constructed by the above-mentioned point and line features,and the minimum plane representation method is used to reduce the dimension of plane representation.A graph optimization model is constructed by the fusing point feature,line feature and plane feature.The objective function which will be optimized is derived and optimal camera pose is calculated iteratively.Finally,aiming at the shortcomings of the Bo W model that ignores the spatial relationship of features and easily leads to perceived confusion,this dissertation design a Bo W model based on comparison of feature space information,and uses the L1 norm distance formula to calculate the vector similarity.Finally,a vision SLAM system by fusing multi feature is designed based on above foundation.The validity and accuracy of the vision location system is verified by the simulation experiment on the data set and the experiment in the real indoor environment.