Research On Indoor Robot Localization Based On Lidar And Vision Information

With the rapid development of sensor technology,communication technology,integrated circuit technology and artificial intelligence technology,the manufacturing cost of the robot has been greatly reduced.As a result,mobile robots have been widely developed in industry,service industry,medical industry,logistics industry and other fields.The field of robotics has gradually become a hot topic in scientific research.The most representative one is indoor mobile robots.The pose perception of indoor mobile robots is a key prerequisite step for tasks such as path planning,obstacle avoidance and navigation.Therefore,indoor robot positioning technology has attracted more and more attention from researchers and technology companies.However,mobile robots are currently facing the problem of poor positioning accuracy by relying on internal motion sensors in complex indoor scenes,and this problem limits the application of mobile robots in indoor occasions,which is urgently needed to be solved.Therefore,this thesis uses vision sensors and 2D Lidar sensors to study the accurate positioning of mobile robots in complex indoor environments.The main contents are as follows:First,this thesis introduces the domestic and foreign research status of mobile robot indoor positioning,and analyzes the advantages and disadvantages of various positioning methods.Then the robot software and hardware platform is introduced,and the principles of the robot motion model and perception model are explained.And according to the needs of the indoor environment in this thesis,different indoor positioning models are established.In this thesis,the odometer track model is used as the input of motion control information.Lidar and vision monocular cameras are used as perceptual sensors,and based on the two perception models.The principles of the two perception models in the positioning process are analyzed.Then,this thesis studies the Monte Carlo positioning algorithm using the odometer motion model and 2D Lidar as the observation model,and introduces the adaptive Monte Carlo positioning that introduces random sampling and Kourbek-Leebler divergence sampling mechanism.algorithm.Experiments verify the feasibility and adaptability of the optimized adaptive Monte Carlo positioning method.On this basis,in order to solve the problem of autonomous relocation after power off and restart during the robot positioning process,a robot relocation method based on particle swarm optimization algorithm is proposed,which introduces automatic relocation based on the traditional particle swarm algorithm.Adapt to the inertial weight and angle search mechanism,improve the local optimization ability of particles and the accuracy of robot relocation.Through experimental verification,the robot relocation method adopted in this thesis is feasible and robust.Next,for the 2D Lidar sensor as the observation model,in the glass wall environment,it is easy to diverge and reduce the positioning accuracy.This thesis proposes an improved artificial road sign visual positioning method based on the visual observation model to improve the indoor positioning of the robot accuracy.Aiming at the problem of the recognition success rate of artificial road sign AprilTag due to uneven indoor light and too fast movement,the traditional AprilTag recognition algorithm is improved,and the image is first grayed out by shifting the tail.And then incorporate the bilinear interpolation down-sampling method to improve the processing speed,and then perform histogram equalization on the processed gray image to solve the problem caused by uneven light.On this basis,the image is bilaterally smoothed and Canny edge detection,enhance image contrast,and eliminate the influence of image noise,so as to improve the success rate of AprilTag image recognition and the positioning accuracy of subsequent work.Finally,the comparison and analysis of the experimental data show that the proposed method has improved the AprilTag image recognition success rate than the traditional method under different light conditions,and the real-time positioning accuracy of the mobile robot during the movement has also been improved,thus verifying The effectiveness and feasibility of the proposed method are verified.Finally,this thesis adopts a fusion sampling method based on hard decision.By processing the Lidar observation information and visual observation information as a binary problem in the sampling process,the visual information is integrated into the adaptive Monte by the threshold decision of choosing one of two.In the Carlo positioning algorithm,the odometer is rectified discretely by visual information to estimate the current position for positioning,which can effectively solve the accuracy of positioning in a glass wall environment.Improve the positioning accuracy of the robot in the glass curtain wall environment.The experimental results prove that the positioning algorithm proposed in this thesis can adapt to the accumulated error of the odometer or the inaccurate positioning caused by the scattering of the Lidar during the positioning process.The adaptability of the algorithm has reached the expected target and verified the effectiveness of the algorithm.