Research Of Augmented Reality Visualization System For Smart Factory Based On Semantic SLAM

With the rapid development of technologies such as the Internet of Things and big data,the traditional visualization methods centered on equipment can no longer meet the production needs of the manufacturing industry,and industrial visualization technology based on augmented reality has become a current research trend.In the field of industrial augmented reality visualization,SLAM technology can realize real-time positioning and augmented reality registration for a wide range of scenarios.However,most of the current augmented reality systems based on SLAM recognize sparse geometric features in the environment,and cannot deal with positioning interference caused by dynamic objects,resulting in poor system robustness and easy tracking loss,which in turn affects production efficiency.To this end,this paper designs and develops a visual augmented reality system for smart factories based on semantic SLAM,proposes an object-level dynamic semantic SLAM algorithm,and constructs an AR scene planning system based on SLAM map reloading.The specific research contents are as follows:(1)Research and implement the object-level SLAM algorithm based on YOLOv5.Based on the traditional visual SLAM algorithm,this algorithm incorporates the semantic information of the object,and represents the object in the real scene as a 3D cube and a dual quadric surface.First,the algorithm uses YOLOv5 object detection algorithm to obtain the semantic information of objects in a single frame image;then,based on the multi-view detection of objects,3D cube proposals and pairwise quadratic surface parameters are generated;finally,3D cubes and pairwise quadratic surfaces are mapped into sparse point cloud maps to construct object-level oriented 3D semantic maps.(2)A SLAM algorithm for high dynamic scenes is proposed.The algorithm first detects the feature points in the camera image frame and the 2D bounding box of the object,and uses the Grab Cut algorithm to segment the object mask in the bounding box;then combines the multi-view geometry algorithm on this basis to detect the potential dynamic objects in the environment,while rejecting all the feature points in the dynamic object mask;finally,the algorithm is compared with ORB-SLAM2 in the TUM public dataset and real scenes for Experimental comparison.The experimental results show that the dynamic SLAM algorithm proposed in this paper has significantly improved the localization accuracy under high dynamic environment.Compared with ORB-SLAM2,the algorithm in this paper reduces the root mean square error,mean error and standard deviation of the high dynamic sequence w＿xyz in the TUM dataset by 97.86%,97.89% and 97.81% respectively,which can effectively eliminate the location interference caused by the high dynamic environment.(3)An AR scene planning system based on SLAM map reloading is constructed.In this system,AR scene planning software,script analysis and AR display software are designed and developed.Firstly,The AR scenario planning software constructs the scenario SLAM map based on the semantic SLAM algorithm in this paper;Then,the RANSAC algorithm is used to fit the plane parameters in the 3D point cloud,and the3 D model is rendered on the plane based on the 3D graphics rendering library Open Scene Graph;Finally,the planned SLAM map and 3D model parameters are saved as intermediate script files.The script analysis and AR display software loads and parses the intermediate script file,obtains the SLAM map and AR registration parameters,and estimates the pose through the relocation module.Meanwhile,the 3D model is placed according to the AR registration parameters to realize the reloading display of the AR scene.Finally,the effectiveness and feasibility of the system is proved by experiments.This paper studies and implements the object-level SLAM algorithm based on YOLOv5,proposes a SLAM algorithm for high dynamic scenes,and constructs an AR scene planning system based on SLAM map reloading based on the improved algorithm,which has certain significance for realizing human-centered industrial data visualization and promoting the development of intelligent manufacturing.