Research On The Method Of Building Semantic Map For Indoor Environment

The completion of various intelligent tasks by mobile service robots in complex indoor environments depends on a deeper understanding and modeling of the surrounding environment.And this process can be implemented by forming a semantic map with rich information.For semantic map,we use the robot real-time localization and mapping methods to build a dense three-dimensional map of the surrounding environment,and then use image semantic segmentation method to semantically label the surrounding environment.In the process of construction a semantic map,it mainly is to solve the construction of three-dimensional environment map,the acquisition of scene semantic information and the integration of semantic information and environment map.Due to the problem of the complex and changeable illumination in indoor environment,this affects the accuracy of 3D environment map and the acquisition of semantic information,and then affects the accuracy of semantic map.Therefore,in order to improve the accuracy of semantic map,this article proposes some solutions to the problem of illumination(1)Designs a full convolutional semantic segmentation network combining the RGB information and the depth information of the scene image.The algorithm first uses the semantic segmentation network to obtain the coarse segmentation results,and then uses the conditional random field model that fuses depth information to optimize the semantic segmentation results.The algorithm can effectively complete the semantic segmentation task in the indoor complex lighting environment.(2)Propose a dense three-dimensional environment map construction algorithm for indoor environment.Firstly,the camera pose is estimated by the feature point matching algorithm,and then the point cloud information is transformed to the same coordinate system by the camera pose to obtain the dense three-dimensional point cloud map.Through the experiments of camera pose estimation and 3D environment mapping,the algorithm is proved to be robust and effective in indoor complex lighting environment(3)During the construction of the 3D semantic map,the association of pixels between adjacent images established in the SLAM algorithm is used to integrate the semantic information of the scene and the 3D environment map information.Finally,the generated semantic map is saved in the form of a point cloud map.(4)Set up the hardware environment and software environment of the entire system,then 3D environment map construction algorithm and the 3D semantic map construction algorithm is run in the Ubuntu 16.04 system.Kinect V2 camera is used to complete the three-dimensional environment map and semantic mapping of different scenes,and the experimental result proves the robustness and feasibility of the algorithm in this paper.