Study On Visual SLAM Based Indoor-Robot Mapping And Navigation Algorithms

The current research of indoor mobile robots mainly focuses on two aspects:construction and positioning.For indoor mobile robots,the construction and positioning functions are differentiated and complement each other.SLAM(Simultaneous Location and Mapping)technology is the key to accomplish these two tasks.The current SLAM technology is divided into laser SLAM and visual SLAM.However,both laser SLAM and visual SLAM have the defects of low operating frequency and large operating resource occupancy.Therefore,it is necessary to integrate with other sensors to obtain better pose estimation and higher operating efficiency.At present,most indoor mobile robots combine inertial measurement unit(IMU)and laser SLAM to perform pose estimation.However,the technology has the disadvantage of relatively high cost and limited installation position of the laser radar.This thesis mainly studies the visual SLAM algorithm based on depth camera,the position and pose estimation algorithm with IMU information fusion and the application of vision-based indoor navigation and positioning algorithm in indoor mobile robots.First,the ORB-SLAM algorithm is studied and improved.The tracking thread in the SLAM algorithm based on multi-sensor fusion is designed,and the 3D point cloud map that can be generated in real time is added.Multi-sensor system includes depth cameras,ultrasonic and motor encoders.During the indoor robot movement,according to the data of the encoder,the new keyframes and map points in the visual SLAM are added to establish a more robust map.Secondly,in view of the low operating frequency of visual SLAM,dependence on the lighting environment,the characteristics of easy-to-lost objects,and the high frequency of IMU's own attitude calculation,which is easy to accumulate measurement errors,this thesis proposes a pose estimation algorithm based on information fusion between visual SLAM and IMU of Kalman filter.Through the fusion of the attitude,the accuracy and frequency of the robot's attitude estimation during the movement are improved.The specific process is:first establish the system equations based on the IMU kinematics model;then perform mathematical modeling based on the deduced error state equations;and finally use the improved ORB-SLAM attitude estimation results as the input of observation equations,based on the Kalman filter principle for system errors the state quantity is updated,thus correcting the posture result calculated by the IMU,so that the robustness of the entire system is improved.Finally,the thesis by using of the point cloud map constructed by visual SLAM,ultrasonic sensors and depth camera sensors,the thesis carries out research experiments on indoor relocation,positioning,and navigation algorithms.Among them,the positioning algorithm relies on the keyframes and feature points saved in the visual SLAM algorithm.Through the feature matching and optimization methods,the re-positioning candidate frames are filtered,so as to complete a certain degree of precision relocation in the indoor environment.Afterwards,an ultrasonic distance sensor is used to assist the depth camera to collect point cloud data,and an artificial potential field method is used to navigate the current position and target position and avoid obstacles in real time.In order to verify the robustness and feasibility of the indoor map building and navigation algorithm,experiments are performed in multiple indoor environments.Experiments in different environments show that the map construction algorithm used in this thesis for information fusion using visual SLAM and IMU can make the construction process more robust and stable.The obstacle avoidance and navigation positioning algorithm using the visual aided ultrasonic sensor can also satisfy the positioning accuracy of the indoor robot using the visual positioning scheme.