Research On SLAM Method Of Indoor Dynamic Scene Based On Semantic Information Fusion

Simultaneous Localization and Mapping(SLAM)is one of the key technologies to realize the active positioning and navigation of modern AGV logistics equipment.Compared with traditional laser SLAM technology,visual SLAM technology has the characteristics of simple structure,low cost and rich information.However,the current visual SLAM system can only achieve stable operation in static scenes,and the system is disturbed in dynamic scenes,which causes the problem of camera tracking estimation deviation and the problem of constructing dense maps.The work of this paper is as follows:First,the FCN semantic segmentation model of this article is constructed based on the full convolutional neural network,and the semantic segmentation model is introduced into the ORB?SLAM2system to realize the semantic information extraction of the input image.Aiming at the problem of camera tracking estimation deviation,a method for optimizing positioning of a dynamic scene SLAM system is proposed.The removal of dynamic feature points is achieved through semantic label images,thereby eliminating the influence of dynamic feature points on pose calculation and improving the accuracy of aircraft pose estimation.Secondly,to solve the problem of constructing dense maps,a static dense map construction method based on semantic information fusion is proposed.Combining semantic information and color clustering search method to achieve the elimination of key frame dynamic point cloud clusters,which improves the construction effect of dense maps in a dynamic environment.Finally,a navigation system based on the visual SLAM system of this article is designed,and the communication between various modules is constructed based on the Robot Operating System platform to achieve the navigation goal.Experimental results show that the system in this paper can accurately identify dynamic targets in the input image and complete semantic segmentation,which can effectively improve the accuracy and robustness of camera tracking,thereby improving the accuracy of system positioning and mapping.In addition,the system can effectively construct a globally consistent static dense map,which can be applied to subsequent navigation tasks to achieve stable operation of navigation tasks.