Research On Autonomous Localization Method Of Indoor Robot Based On Vision SLAM

With the rapid development of modern science and technology,high-tech products continue to enter people's sight.Among them,robots have always been highly concerned by people,which represents the highest achievements of human science and technology.An important component of robots is mobile robots,and its application scenarios are gradually increasing.Among the key technologies involved in mobile robots,an important issue that needs to be solved urgently is robot positioning.Among them,in the environment with weak GPS signals or even without GPS,it is the SLAM problem in robot engineering to realize the autonomous positioning and mapping of mobile robots.Because of the low cost,easy access to information,low power,and rich data acquisition of sensors adopted by Visual SLAM,it has become a hotspot of current research.This article first analyzes the purpose and significance of the vision-based SLAM research,introduces the current mainstream visual SLAM research programs,and uses the stereo vision SLAM method based on ORB features through research and analysis.Secondly,this article studies the basic theory of visual SLAM.The related theoretical basis of camera imaging model is expounded,which mainly introduces the camera's coordinate system,common pinhole cameras and their models,and the problems of distortion caused by camera imaging.Aiming at the camera movement in three-dimensional space,this paper introduces several commonly used methods of camera pose representation and their mutual conversion.Aiming at the problem of mismatch in the front-end feature matching of visual SLAM and the unstable number of iterations of the traditional mismatch elimination method,this paper proposes an improved mismatch elimination algorithm.Through the improved algorithm,the accuracy of camera feature matching is ensured,and a more accurate camera pose can be obtained according to the precise feature matching.Then,non-linear optimization is performed on the camera pose obtained from motion estimation.Several commonly used non-linear optimization methods are introduced,namely the Gauss-Newton method and LevenbergMarquadt method,and the differences between them are analyzed.In order to solve the problem that pure vision SLAM cannot work stably when the camera movement is too fast or the camera operating environment features are insufficient,this paper proposes an indoor positioning algorithm for stereo vision fusion IMU information.Through a comparative analysis of the current tightly coupled and loosely coupled vision and IMU fusion methods,the tightly coupled fusion method is used to fuse the IMU information.Aiming at the problem of inconsistent sampling frequency of IMU information and visual information,this article uses pre-integration method to process IMU information,and simultaneously establishes accelerometer and gyroscope noise models.Through the analysis of IMU error propagation and bias analysis,it provides a prerequisite for the fusion of vision and IMU.In order to evaluate the effectiveness of the algorithm proposed in this paper,this paper first calibrates the stereo camera used in the experiment,and jointly calibrates the camera and IMU according to the obtained camera internal parameters.Then design and build the software and hardware platform needed for the experiment to perform indoor robot positioning experiment,and perform the optimized false matching rejection algorithm experiment and visual fusion IMU algorithm experiment.The experimental results show that the optimized feature matching algorithm is superior to the traditional algorithm.In the robot localization experiment in the actual corridor environment,the number of key frames and landmark points in the surrounding environment extracted by the improved algorithm is greater than the traditional algorithm.The vision fusion IMU algorithm has less environment features and the positioning accuracy of the robot at high speed is greater than that of pure visual positioning.