| All kinds of robots play an increasingly important role in the life and production activities of modern society.And the ability to perceive the environment is an important basis for robots to perform various tasks.Visual simultaneous positioning and mapping technology is one of the key technologies for mobile robots to realize environmental perception,and it is also a research hotspot in the field of robotics in recent years.But improving the robustness of visual SLAM techniques to the environment is still a challenging work.This paper aims to improve the positioning accuracy and robustness of mobile robots in indoor environments.Aiming at solving the problems of visual positioning in weak texture environments,the decrease of visual positioning accuracy caused by illumination changes,and the vulnerability of pure visual synchronous positioning and mapping to environmental factors,the research on visual SLAM method based on edge alignment,online photometric calibration,visual-inertial combined positioning,etc.is carried out.The main work of this paper is as follows:1)Aiming at the visual localization problem in the weak texture environment,the research on the visual SLAM method based on edge alignment is carried out.In this part,the basic principles of edge extraction and distance transformation algorithms are explored.On this basis,a detailed theoretical derivation is given for both camera pose estimation based on edge alignment and back-end sliding window optimization.A loop closure detection algorithm based on random ferns is used to eliminate accumulated error.The proposed method based on edge alignment achieves effective localization in weak texture environments,and the root mean square error of the translation part of the absolute trajectory is 0.09 m.2)Aiming at the problem that the accuracy of visual positioning decreases due to illumination changes,the research on the edge-aligned vision SLAM method integrated with online photometric calibration is carried out.In this part,the image photometric modeling and photometric calibration algorithm were explored.On this basis,a visual SLAM method integrating photometric calibration was proposed.The specific application of M estimation in SLAM is briefly discussed,and the T-distribution weight function is used to improve the stability of the algorithm for multiple runs.On the TUM-RGBD dataset,it is verified that the integrated online photometric can effectively improve the positioning accuracy,and the error is reduced by about 56.2% on average compared with the visual SLAM method based on edge alignment.3)Aiming at the low robustness of visual SLAM methods to the environment and motion,the research on visual SLAM methods based on edge alignment and inertial assistance is carried out.In this part,the IMU pre-integration,gravity direction initialization,and dynamic marginalization are explored,and the detailed theoretical derivation of camera pose estimation and back-end optimization problems in visual-inertial SLAM based on edge alignment is given.On this basis,a visual-inertial SLAM method based on edge alignment is proposed.The performance advantage of this method compared with the pure vision method is verified by the dataset,and the localization error is reduced by about 53.54% on average compared with the method based on edge alignment.4)A prototype experimental platform was built.The performance of all the above algorithms is verified on the experimental platform.Several groups of experiments,such as weak texture environment experiment,illumination change experiment,visual-inertial combination experiment,and comprehensive positioning experiment,were designed.The evaluation is carried out with the trajectory of the lidar as the reference ground truth. |
