Research On Vision-based SLAM And Autonomous Navigation For Mobile Robot

Localization,mapping and navigation are the three key technologies for mobile robots to achieve its autonomous mobility.Furthermore,Simultaneous Localization and Mapping(SLAM)technology is the basis of applied researches,like navigation and mission planning,which is the precondition to realize the robot autonomy.Currently,the dominant SLAM method is divided into two parts,namely,Laser SLAM and Visual SLAM.Visual SLAM has attracted more and more attention in the field of SLAM owing to its low price,compact size and the ability to obtain abundant scene information of visual sensors.The point-based Visual SLAM method represented by ORB-SLAM2 still has shortcomings in practical applications,such as low accuracy of localization and can not be directly used for navigation.This thesis proposes some improvement schemes on the basis of ORB-SLAM2 with the purpose of improving the accuracy and robustness of localization and mapping for Visual SLAM system.At the same time,a set of mobile robot autonomous navigation system based on Visual SLAM is implemented in ROS(Robot Operating System).The main research work of this thesis are as follows:(1)A stereo Visual SLAM system based on point-line features is established.Aiming at the defect that low localization accuracy of the point-based Visual SLAM method in low-texture environment,a stereo Visual SLAM method fusing points and line segments is proposed on the basis of point-based Visual SLAM method,which avoid the decrease in camera localization accuracy caused by the lack of point features in low-texture scenes.After the corresponding features are found in different frames,the motion of the camera is estimated through minimizing the re-projection error of the points and line segments.Finally,the experimental results illustrate that the localization accuracy of the proposed method is superior to the point-based Visual SLAM in most dataset sequences.(2)A localization method via fusing wheel encoder and stereo vision is presented.In order to avoid the disadvantages of pure Visual SLAM under actual circumstances,such as being sensitive to the change of environment,illumination and motion speed,a localization system through combining the wheel encoder and stereo vision based on the graph optimization model is proposed.The localization method utilizes the wheel encoder that can provide high precise localization in short time to convert the constraints in robot coordinate to the camera coordinate,which can further assist feature tracking and constrain the camera poses in different moments.The camera motion is achieved through odometer motion error together with re-projection error,which can further improve the accuracy and robustness of location and mapping.(3)The autonomous navigation of mobile robot based on Visual SLAM in unknown scene is realized.The sparse feature map used for relocalization and the occupied grid map applied for path planning are constructed through the improved Visual SLAM method together with a depth sensor at the same time.The relocalization of the robot is completed by feature matching in loaded sparse feature map,then the path planning between current location and target position is carried out by path planning algorithm.Finally the practicability of the occupied grid map construction and autonomous navigation system are proved in real robot platform.