Design And Implementation Of Indoor Visual SLAM Based On Depth Image Information

With the development of the industry and intelligent logistics,AGV robots were widely used.The core technology of AGV robot was autonomous positioning.However,in the environment where indoor GPS signal was lost,positioning failure might occur.Its positioning and navigation function needed a priori information to preset the environment,and the actual map information couldn't be updated in real time when the environment changed.This led to the simultaneous visual localization and mapping(visual SLAM)technology,which could effectively solve above problems.The design of this dissertation aimed at fusing the data and information of deep vision sensor and Inertial Measurement Unit(IMU)to achieve map construction in an unknown environment.The following works were completed:(1)In view of problems like the lack of features in the indoor environment,insufficient lighting,long system initialization time of the traditional monocular SLAM,and uncertain scale,this dissertation designed a visual SLAM solution based on RGD-D camera,including ORB feature point extraction and matching on the front-end,improved PL-ICP pose estimation algorithm,and back-end graph optimization algorithm.Compared with the traditional monocular SLAM,errors were significantly reduced.(2)In order to improve the positioning accuracy and robustness of deep visual SLAM,this dissertation further designed a tight coupling of deep visual information and IMU data,constructing a least square error objective function based on the measurement data of the sensor and giving a pose solution method.The SLAM fused with deep vision inertial navigation was run on an Eu Ro C public data set.Compared with the deep vision SLAM experimental results,all data of the SLAM fused with deep vision inertial navigation were better than that of pure deep vision SLAM,and the RMSE was reduced from 0.248 m to 0.043 m.(3)In order to complete the experimental algorithm verification on the robot platform,the ROS-based AGV robot model and simulation environment were built,and the accurate navigation raster map is constructed.And the calibration experiment of the actual hardware sensor was further implemented.The experiment finally proved that the SLAM fused with deep vision inertial navigation designed in this dissertation improved the positioning accuracy and robustness and met the needs of engineering design.