Research On SLAM Technology Based On Semantic Segmentation And Multi View Geometry In Dynamic Scene

Mobile robots are widely used in storage,automatic cleaning,cargo transportation and other scenes,while real-time localization and mapping are the basis of robot's perception of environmental information and autonomous navigation.This paper studies the localization and environment reconstruction of indoor mobile robot in dynamic scene.The purpose of this paper is to make use of semantic information so that the robot can better locate in the dynamic scene,and build the map of the dynamic scene.The detection of moving objects in the scene is the key technology of dynamic slam.The traditional visual SLAM algorithm is based on the assumption of static environment.The interference of dynamic objects will cause great errors in localization accuracy.In order to solve this problem,this paper uses the method of semantic segmentation and multi view geometry to detect the dynamic object area of the image,and proves the effectiveness of this algorithm through experiments.Accurate attitude estimation is the basis of mobile robot's specific work in unknown environment.In order to make the mobile robot get accurate attitude in real time in dynamic scene,this paper designs a SLAM algorithm which is suitable for dynamic scene,integrates dynamic object segmentation into visual slam,and compares the improved algorithm with the original algorithm in the open data set.Experimental results show that the improved method can effectively reduce the location error of visual slam in dynamic scene.Building scene map is another important work of vision slam,and it is also the basis of navigation and obstacle avoidance for mobile robot.In this paper,the method of building environment map based on scene segmentation is used,and a map segmentation method based on plane fitting is used to divide the map point cloud into ground point cloud and non ground point cloud,and the octree map containing non ground information is constructed.In addition,an object tracking based point cloud ghosting method is proposed.The proposed algorithm can effectively improve the map quality and facilitate the robot navigation.This paper optimizes the localization accuracy and map effect of visual slam in dynamic scene.The recognition of dynamic objects can improve the localization accuracy of vision slam system,and the scheme of building map is also a good idea of building environment map,which provides a reference scheme for vision slam technology applied to indoor mobile service robot.