Algorithm Implementation Of Simultaneous Localization And Mapping Of Mobile Robot Based On Vision

Simultaneous Localization and Mapping(SLAM)completes the autonomous locating and mapping in the unknown environment through the sensors,which is the key to the realization of intelligent robots.Autonomous locating is mainly divided into laser SLAM and visual SLAM.The scene information obtained by the expensive laser radar is single.The visual sensor can obtain rich scene information,but the traditional visual SLAM has the problem that it is easy lost in low-texture and fast motion.Therefore,this thesis research an improved visual SLAM based on the mobile robot platform.This main researches include: 1.The Inertial Measurement Unit(IMU)information is integrated which based on the framework of the pure visual SLAM system,and the IMU pre-integrated value is used as the estimated pose of the current image in the pose tracking process,which makes it more in line with the actual motion model.Compared with the traditional pose optimization of visual SLAM,the scale constraint of stereo visual and the IMU motion constraint between keyframe and keyframe are increased,which improves the accuracy and robustness of the system.2.This thesis presents a loop detection method based on improved auto-encoding network,which to predict the feature vector of the current image by using the trained network model,and retrieves the closure frame in the feature space by calculating the similarity of feature vectors,and then to eliminate the error throng their relative pose.3.This thesis implements the preservation of point cloud maps and the capability of localization by loading the offline map,which avoid the repeated mapping in the same scene.This thesis implemented an improved stereo visual SLAM algorithm.By comparing the locating accuracy test between the proposed algorithm and the traditional algorithm on the 11 sub-datasets of EUROC,this proposed algorithm has 7 sub-datasets showed the best performance.The accuracy of the loop closure is tested by the dataset.When the recall rate is within 60%,the accuracy of the improved algorithm is more than 75%,and the retrieval speed of the closure frame is increased by 34% compared with the traditional algorithm.In the same scene,offline map-based locating effectively saves 13% of computing resources compared to simultaneous locating and mapping.The feasibility of the proposed algorithm is proved by testing the algorithm in the actual scene on mobile robot.The improved stereo visual SLAM algorithm proposed in this thesis is better than traditional algorithms in the case of real-time performance.It can be applied to scenes such as sweeping machine,driverless and augmented reality.So,the proposed algorithm has an application prospects.