Collaborative Visual 3D Information Estimation For Natural Scenes

The 3d information estimation technology based on sparse SLAM(Simultaneous Localization and Mapping)has important research significance and application value in 3D-Reconstruction,Path Planning,AR/VR and other fields.However,at present,the system is often built based on single individuals,and it is unable to efficiently and comprehensively estimate the 3d information of the scene when dealing with largescale and varied natural scenes.To achieve a rapid and comprehensive estimation of 3d information in natural scenes,this paper proposes a multi-agent collaborative SLAM framework estimation method,which can quickly estimate 3d information in large scenes.The main research contents and achievements are as follows:1.A aerial-ground collaborative sparse SLAM framework is proposed.The framework can quickly generate global sparse map for large scenes with the help of multi-view information.In the study of the image feature detection algorithm,to solve the problem of quick estimate three-dimensional information,this paper adopts the ORB(Oriented FAST and Rotated BRIEF)features as a foundation for the collaborative framework.Finally,the pose of the unmanned ground vehicles(UVGs)and unmanned aerial vehicles(UAVs)is optimized based on the Marker.The experimental results show that the proposed method can be used to quickly estimate the 3d information of the scene by combining the ground end and the air end.2.A marker-based method of map scale estimation and map alignment is proposed,which can fuse the local map of agents and recover the actual scale of 3d information with the help of the prior information of marker and pose information.This method can estimate its pose information by detecting the previous marker.Then the positions and scales of the marker are used as the prior information to calculate the distance between the marker center and the observed agent center.Finally,the redundant points in the map integration are removed,and the local maps are aligned and merged into the global map.Experimental results show that the proposed method can obtain 3d information on a real scale.3.Set up the volume estimation system of the industrial pile.Based on the framework of aerial-ground collaborative sparse SLAM,the system can use the image sequence captured by UAV and UGV to calculate 3d information,identify the target pile in the reconstruction point cloud,and finally estimate the volume of the target pile.The experimental results show that the proposed method is feasible for industrial applications and can provide high precision pile estimation results in a short time.In the end,this paper summarizes the work done and proposes further work objectives,including using GPU to accelerate the algorithm,adding multiple cameras and IMU equipment to improve the robustness.