Research On Simultaneous Localization And Semantic Mapping Of Indoor Dynamic Scenes

With the rapid development of new technologies such as computer vision,artificial intelligence and 5G,Simultaneous Localization and Mapping(SLAM),as a key technology in the field of artificial intelligence applications such as driverless,mobile robots,has become a new research hotspot.The current research on SLAM is mostly based on the assumption of static scenes,while the dynamic objects in the indoor environment are inevitable.The assumption based on static scenes greatly limits the development of SLAM and the application of SLAM system in real life.Moreover,SLAM system without semantic information maps cannot satisfy the perception of the environment such as unmanned vehicles and mobile robots.Therefore,based on the current research on the development status of SLAM at home and abroad and the in-depth analysis of the existing technical framework,this paper adds the semantic segmentation network to the visual SLAM system and proposes a new feature point screening method.The screening method has formed a new visual SLAM system in dynamic scenes.In this paper,SLAM is studied from semantic and dynamic scenarios.The specific research contents are summarized as follows:(1)The PSPNet semantic segmentation network is added to SLAM to achieve Simultaneous Localization and Semantic Mapping,which is semantic SLAM.The semantic segmentation network associates images with semantics and combines with SLAM system to generate a semantic map containing environmental information.It gives mobile carriers such as mobile robots high-level perception capabilities for mobile carriers to perform cognitive and task reasoning,and to improve robot service capabilities and human-computer interaction intelligence.Experiments show that the improved method can generate semantic point cloud map and semantic octree map with semantic information,which reflects the feasibility of the system proposed in the paper.(2)An optical flow and semantic segmentation method is proposed to filter dynamic points.Aiming at the problem that the dynamic object in the dynamic scene affects the camera pose estimation,causes the track error to be too large,and even causes the system to collapse,a method of secondary selection of feature points based on optical flow and semantic segmentation is proposed.The results show that the secondary screening can effectively reduce the trajectory error and drift in dynamic scene,and improve the stability of slam system in dynamic scene.