Design Of Autonomous Positioning System For Robot Based On Monocular Vision And Multi-sensor Fusion

With the advent of the era of intelligent manufacturing,the development of robot technology has received widespread attention.Intelligent mobile robots can be used in industry and service industries,saving a lot of economic and human costs and enriching people's lives.When intelligent mobile robots perform various tasks,accurate positioning is the first problem to be solved,so it has very important research significance.Based on the self-developed robot formation and target rounding experimental platform,this paper uses the monocular camera,encoder and IMU on the robot to design and implement a multi-sensor-based robot autonomous positioning system.At the same time,the operation and positioning are reused in one camera,which saves hardware costs and computing costs,and has high positioning accuracy,which is suitable for indoor and outdoor high-precision mixed formation operations.The main work of this article is as follows:(1)This article researches and applies common feature extraction algorithms,including KAZE algorithm,FAST algorithm,ORB algorithm and BRISK algorithm.Then the article researches and applies feature point matching methods,including feature matching and optical flow tracking.In the laboratory scenario,the camera equipment carried by the robot is used to collect images,and detailed experiments are designed to compare various algorithms.Finally,it is determined to use FAST feature extraction algorithm and optical flow tracking algorithm to perform image feature extraction and pairing,respectively.(2)In the scene where the operation and positioning are performed simultaneously,a large number of dynamic targets related to the task will appear in the image originally used for positioning,and these dynamic targets will cause great interference to the positioning of the monocular visual odometer.In response to this problem,a static background feature point screening algorithm is designed in this paper to effectively remove the dynamic target feature points in the image.First,an algorithm for matching feature points of image sequences based on IMU information is proposed.Then,a feature point classification method based on cluster analysis is designed.Finally,the dynamic target feature points are removed according to the feature point classification results.(3)Aiming at the problem that the camera pose estimated by the traditional monocular visual odometer method has scale uncertainty,an improved monocular visual odometer method is proposed,so that the motion estimation of the monocular camera can be applied to autonomous positioning of the mobile robot.First,the data of IMU sensor and encoder are fused by EKF algorithm to obtain the translation vector between image frames.Then the traditional monocular visual odometer method is improved by combining 2D-2D and 3D-2D camera motion estimation methods,and the 3D points are screened twice to further obtain the optimal 3D points.(4)The validity and applicability of the proposed algorithm are verified by experiments in data set and indoor environment.By analyzing the positioning error,it is verified that the proposed method can realize the precise positioning of the mobile robot and meet the needs of the project.