The Method Of Binocular Vision-SLAM For Indoor Robot

Family service robots attract more and more attention in society,with the increasing of aging population.Simultaneous Localization and Mapping(SLAM)is the key to achieve the independent navigation for mobile robot.The SLAM called is the process of building environment map based on robot own estimate position and pose and environment perception from sensor,at the same time using the map for localization.In recent years,as the progress of digital image processing technology,vision SLAM gradually become the research hot in the field of mobile robot.In this thesis,a binocular vision SLAM method is researched.The specific content as follows:Firstly,for the problem of acquiring binocular vision depth information on a mobile robot platform,a binocular image stereo matching algorithm which is applicable to dynamic image sequence is researched.First,with a mature Semi-Global Matching algorithm,the rapid stereo matches of stereo images are achieved to obtain the disparity images.On this basis,the major noises encountered during carrying out the stereo matches on a mobile platform in a complex environment are analyzed,including the white-spot noise as well as the fault flicker noise.Further,a white-spot filtering algorithm and a disparity image time-domain filtering algorithm are proposed respectively to reject these two noises.At last,the experimental verifications are performed,and the results indicate that the noises coming from stereo matches of dynamic binocular vision images are effectively suppressed by the proposed algorithms,besides a continuous and clear disparity image sequence can be derived in real-time,which is significant to the obstacle avoidance and the situational awareness of a running mobile robot in a complicated environment.Next,a data associated method is researched,data association is key to localization,the false data association may lead to the divergence of the process of SLAM.In this thesis,mutual matching between current feature points and the feature points in the map,increased the accuracy of data association.Then,aimed at enormous computation cost of EKF algorithm,it is difficult to solve the SLAM online,a SLAM method of separating the environment-learning from localization was researched.Based on original algorithm of EKF-SLAM,multi-progress was proposed,including the thought of separating learning and localization,fold-line-SLAM method,thereby enormous computation cost was avoided.Finally,based on mobile platform and binocular stereo cameras in our laboratory,thefold-line SLAM environment learning,online localizing and collision avoidance and path planning experiments.The results indicate the effectiveness and practicability of the relevant algorithms proposed.