Visual-Fusion SLAM Technology And Its Application In Mobile Robots

SLAM(Simultaneous Localization and Mapping),that is,simultaneous localization and mapping,refers to a robot carrying a specific sensor,which builds a map and estimates its position and posture in the course of motion without prior environmental information.When the sensor on board is a camera,it is called Visual SLAM.SLAM technology is the key and core of autonomous navigation of robots,which is called the eyes of robots.It has important applications in the fields of automatic driving,three-dimensional mapping and AR(Augmented Reality).After years of development,visual SLAM technology with vision sensor as the core has become a research hotspot.Among them,monocular vision SLAM is most popular because of its low cost and simple application.However,monocular vision SLAM lacks scale information and can not work in complex environments,which requires other sensors to complement each other.Therefore,multi-sensor fusion SLAM will be a hot research topic in the future.In this paper,the fusion of monocular vision with inertia,magnetic,lidar and other sensors is studied,and it is applied to the inspection of foreign bodies on the road surface of robots.In order to verify the method in the actual environment,an omnidirectional robot mobile platform is built.Specific research work is as follows:(1)A SLAM method based on loosely coupled fusion of inertial/magnetic sensors and monocular vision is proposed.To solve the problem that the monocular vision SLAM algorithm lacks the scale information and can't be used when the camera moves too fast,a SLAM method based on loose coupling fusion of inertial/magnetic sensors and monocular vision is proposed.Firstly,a fuzzy adaptive nine-axis attitude fusion algorithm is improved to estimate the heading angle of IMU(inertial measurement unit)with high accuracy.Then,a monocular ORB-SLAM2(oriented FAST and rotated BRIEF SLAM2)algorithm is used,its scale factor is calculated by IMU,and scale transformation of its output pose information is performed.Finally,the loosely coupled method is adopted to fuse the poses of IMU and ORB-SLAM2 algorithm after scale transformation by Kalman filter.The experiments are carried out on a public dataset EuRoC,and the results show that the total root mean square error of the method is 5.7 cm.For further verification in the actual environment,an omnidirectional mobile platform is designed and built,and the test results show that the rotation angle error of the method is less than 5 degrees and the total position RMSE is 9.8 cm compared with the pose data measurements from the lidar.(2)Visual-fusion SLAM technology is applied to the foreign body detection on the road surface of the robot.In order to meet the requirement of all-weather road detection,a foreign body detection algorithm based on multi-line lidar and assisted by viusal is proposed.Laser radar data is corrected by visual fusion of SLAM system output information such as pose and speed,so that the X-axis direction of point cloud data is parallel to the road direction,so that the point cloud of road and object on the road can be quickly segmented and preprocessed.For the pre-processed point clouds,a number of plane models are used to fit the ground to realize the fast segmentation between the point clouds on the road and the point clouds on the object;the point clouds on the object are clustered by the neighborhood clustering algorithm and the relative position and size of the object are output;after clustering,the model is simplified and the speed is estimated by the method of data association;and then the estimated relative speed and position are transformed into the world coordinate system.Finally,in order to ensure the robustness of the algorithm,the position and velocity of the object in the world coordinate system are used to make double judgments.The data packages collected by the 16-line lidar,the road inside campus and the road outside campus were tested respectively.The test results show that the foreign body detection method proposed in this paper can eliminate normal driving vehicles without false detection,and can effectively detect static foreign bodies with a height greater than 40 cm.(3)Design and system implementation of omnidirectional mobile robot platform.In order to test and apply the proposed method in the actual scene,an omnidirectional mobile robot platform is designed and built,and its kinematics is analyzed.Mobile robot platform is mainly equipped with monocular camera,three-dimensional lidar,two-dimensional lidar,IMU and other sensors.The main algorithm is developed based on ROS(robot operation system),and runs on PC as decision-making layer.The feasibility of SLAM method based on the fusion of inertial/magnetic sensors and monocular vision and the detection method of foreign body in pavement based on the fusion of vision SLAM technology are verified in practical environment.