Research Of Feature Matching And Location In Binocular Stereo Vision SLAM

Simultaneous location and mapping(SLAM)is a process by which a mobile robot can build an environment map and compute the location at the same time.Solving SLAM problem have great significance in realizing mobile rebot self-location,and it has also became hot topic in mobile robot navigation and computer vision.Because of the mismatch and high complexity in SLAM,which lead to long period of building the map and poor real-time performance.Therefore,we studied the feature matching and location of SLAM based on theoretical research.In this paper,the binocular stereo vision SLAM system is deeply studied.The main work is as follows:Firstly,we introduce the systematic framework of SLAM and analyze two common used SLAM filter algorithms that is Kalman Filter and Particle Filter.The paper also give the observation and motion model of binocular stereo vision SLAM based on which a systematic framework of binocular vision SLAM is established.Secondly,the data association of binocular vision SLAM is studied.First,we introduce the knownology of feature extraction and feature matching.Then,in order to solve SIFT high dimension and large mismatch rate,SMO-SIFT algorithm is proposed which combine SMO to SIFT.The MATLAB simulation proves that SMO-SIFT alrgorthm performs well on reducing dimension,improving real-time and increasing accuracy.Thirdly,the path estimation of SLAM is studied.First,we introduce the RaoBlackwellised particle filter(RBPF).Then,in order to solve RBPF behavior of frequent resampling results in “particle impoverishment” problem,INGO-RBPF is proposed which is based on improved niched genetic optimization(INGO)algorithm.The MATLAB simulation proves that INGO-RBPF performs well on estimated accuracy,stability,disturbance and location accuracy,and therefore it is suitable to apply in SLAM real-time location.Finally,we apply the SMO-SIFT and INGO-RBPF algorithm to the robot operating system(ROS)in experiment environment.We give software and hardware design of three modules: map,robotics and remote control.The experiment proves that robot can build map and location successfully which shows the result is desirable.