Research And Implementation Of SLAM Algorithm Based On Stereo Cameras And IMU

With the rapid development of artificial intelligence technology and the needs of various industries in society,robots capable of autonomous mobile navigation and obstacle avoidance have received widespread attention and use.Simultaneous Localization and Mapping(SLAM)is one of the core technologies of autonomous mobile robots.Research on SLAM technology is of great significance to the development and application of robots.Visual sensors are widely used in the field of indoor mobile robots because of their cost-effective and rich image information.However,SLAM based on only visual is over dependent on the surrounding environmental conditions and cannot handle dynamic scenes and scenes with missing texture features.It is seriously affected by lighting,rdinary cameras cannot deal with the rapid carrier movement due to the low frame rate.The Inertial Measurement Unit(IMU)can measure the acceleration and angular velocity of the carrier in real time,and has good complementarity with the camera.Based on this,this thesis proposes a visual inertial navigation SLAM system based on the fusion of stereo visual and IMU.The front-end data preprocessing and back-end optimization of the visual inertial navigation SLAM system based on stereo visual and IMU are studied.For the front end,the camera uses FAST feature extraction and uses the optical flow method to track the features.Based on the pre-integration theory,the IMU solves the sensor's speed,displacement,pose and IMU offset for back-end optimization.The back end builds a nonlinear optimization system based on sliding windows,and uses tightly coupled methods to fuse the information of the camera and the IMU to optimize their error terms simultaneously.The loop detection and global optimization based on the bag of words model are added to build a complete SLAM system to improve the accuracy of the system and reduce the cumulative error.In addition,the visual inertial navigation SLAM will output the constructed map and realize the function of using the map for relocation.At the end of this thesis,an experiment was carried out on the built stereo visual inertial navigation SLAM system,which proved the feasibility and effectiveness of the scheme.The absolute pose error of the system and the running time of the system are analyzed,and compared with the visual inertial solution OKVIS,and the Xiaomi camera is used to run in the actual environment.The results show that the accuracy of the system proposed in this thesis is slightly better than the OKVIS method,and it has better real-time and robustness.