Research On Indoor Semantic SLAM Algorithm Based On RGB-D Camera

Visual SLAM(Simultaneous Localization And Mapping)is one of the core technologies for mobile robots to perceive the surrounding environment and locate,and has become a research hotspot in the field of intelligent robot.Aiming at the problems of traditional visual SLAM being sensitive to lighting changes in indoor environments and the lack of semantic information in the generated three-dimensional maps,this paper studies a highly robust semantic vision SLAM algorithm based on RGB-D cameras for the autonomous navigation function of mobile robots.The algorithm can realizes the motion positioning and map construction of the three-dimensional environment by sensing the semantic information of the environment,and has high visual SLAM positioning accuracy and semantic map construction robustness.The main work and conclusions of this paper are as follows:1.Aiming at the problems of visual navigation of autonomous mobile robots in indoor environments,such as lack of texture,fluctuations in the mobile platform,and significant illumination changes,a multi-feature point fusion Visual odometer method was designed to improve the positioning accuracy and robustness of the visual SLAM algorithm.The improved algorithm combines the ORB feature detection algorithm with the BRISK feature detection algorithm,which has strong robustness to light and blur information,so that higher quality feature points can be obtained to estimate the camera pose after the improvement.Experimental results show that the improved mileage calculation method can effectively speed up the extraction and matching of feature points,optimize the camera pose estimation value,improve the robustness of the SLAM algorithm,and facilitate the generation of high-quality 3D point clouds.2.The YOLOv5 target detection algorithm is used to acquire semantic information of indoor scenes.First,make the corresponding data set by manually taking pictures,then mark and train the target detection network model,and finally verify the accuracy of target detection through experiments,and integrate the trained YOLOv5 algorithm into the visual SLAM algorithm.3.A fast point cloud segmentation algorithm is Proposed based on color and geometric information.the hypervoxel clustering algorithm is used to preprocess the original point cloud,and a color and geometric information hybrid metric is used to refine and merge the point cloud data over-segmented by the traditional algorithm,reducing the complexity of subsequent calculations;combining the local part of the point cloud space Convex information and global plane information,a segmentation algorithm is used based on graph theory,which finds the smallest energy segmentation line of the point cloud voxel adjacent graph,and performs segmentation,and builds a three-dimensional map.Experiments shows that the improved algorithm has a great improvement in segmentation effect and segmentation speed.4.Combine geometric information and semantic information to design a three-dimensional semantic map generation algorithm based on the YOLOv5 target detection algorithm.The image has rich texture and color information,and the three-dimensional point cloud data also has rich spatial geometric information.This paper uses the deep learning target detection algorithm to process the environmental RGB image to extract semantic information,and then segment the reconstructed point cloud.Finally,the two The integration of the authors realizes the construction of point cloud semantic map.