Construction And Research Of Multi-Modal Fusion Semantic SLAM System In Indoor Environment

Simultaneous localization and Mapping(SLAM)technology can accurately locate and map the environment according to its own sensor information,which is widely used in unmanned delivery and automatic driving scenes.Laser and visual SLAM using a single sensor have some limitations in dealing with complex environments.In this paper,a multi-mode fusion semantic SLAM system combining semantic information for loopback detection is constructed.The main research contents are as follows:(1)A multiscale multipath attention semantic segmentation network is constructed.The network structure is mainly divided into three paths,namely pyramid spatial path,context path,semantic graph path.The pyramid spatial path encodes the multi-scale structure and spatial location information of the image,the context path extracts the context information of the image,and the semantic graph path extracts the more abstract high-level semantic information by the way of graph convolution.Fast and accurate semantic segmentation is achieved by combining multi-scale information,multi-path structure,and attention calculation.(2)A multi-modal fusion semantic SLAM system is constructed.The camera,inertial sensor and lidar sensor are used to obtain the environment information,the visualinertial odometry and laser-inertial odometry is established to estimate the position of the mobile platform.Use the global position factor graph optimization for more robust and accurate position information.Using semantic segmentation add visual loop constraints,to correct the system accumulated error.(3)Set up experimental platform for verification experiment.In order to verify the feasibility and practicability of the proposed system,a set of autonomous mobile platform is built based on the four-wheel differential chassis model.With the help of the built autonomous mobile platform,this paper carried out the synchronous positioning and mapping experiment of the autonomous mobile platform in an indoor environment,and carried out the navigation experiment of the autonomous mobile platform based on the constructed map.The experimental results show that the proposed semantic SLAM multimodal fusion system has feasibility and practicability.Finally,this paper summarizes the proposed system in detail and gives a simple outlook on the direction of improvement.