Design And Implementation Of Mobile Robot Positioning Algorithm Based On Visual Odometry

With more and more applications of mobile robots,the autonomous behavior and decision making of mobile robots in complex environments have always been one of the hot spots.The positioning and navigation of robots are indispensable parts.The traditional robot positioning uses the combination of inertial navigation and GPS.However,there are data loss and congestion in the GPS,and it also has to be limited in outdoor scenes and other conditions,which has certain limitations.Visual positioning does not have the above problems.Visual positioning,also known as visual odometry,refers to the extraction of image information obtained by a camera during the operation of a mobile robot,and estimates the position of the robot.This dissertation focuses on depth camera-based visual odometry method and proposes a visual odometer method that combines Kalman fusion with direct optical flow method and feature point matching.The specific content is as follows:First,several common feature value extraction and matching algorithms are analyzed.The SURF algorithm is used to process the key frame image.Then,the motion estimation of the optical flow method is analyzed,and put forward some measures to improve the accuracy of optical flow odometer.The feature point matching method and the optical flow method are used for Kalman fusion and the idea of local maps for key frames was proposed.The data set is used for testing the correctness of the fusion algorithm.Finally,an experimental framework for visual fusion algorithm is designed and implemented.It is implemented in real scenes under different operating conditions,and the real-time and robustness of the algorithm are verified.The results show that the fusion algorithm can overcome the shortcomings of the optical flow and the real-time processing speed of the feature point method.It highlights the advantages of high accuracy of feature point matching and cumulative error elimination of local maps.This method can provide accurate real-time positioning information for laser SLAM with a certain degree of robustness.