Research On Robotic 3D SlAM Algorithm In Dynamic Environment

Simultaneous Localization and Mapping(SLAM)is a basic problem and research hotspot in robot research.Data association and dynamic target processing are the key issues in the application of SLAM technology in a dynamic environment.It is an important prerequisite for applying SLAM technology in areas such as driverless,home services,and augmented reality.In this paper,Bumblebee camera and Turtleble mobile base are taken as the experimental platform to study data association and dynamic target processing.Three algorithms are proposed for classification of planar dynamic environment features,classification of non-planar dynamic environment features and initialization of dynamic environment maps,and three algorithms are proposed.Based on the fusion texture and region constraint algorithm,the 3D SLAM algorithm in dynamic environment is proposed.The specific research content is as follows:Firstly,a plane dynamic environment feature classification algorithm is proposed for the problem of target detection(different speed and different size)under plane dynamic environment.Through the re-projection error algorithm,the initial classification of feature points is implemented,and the texture detection algorithm and the region constraint algorithm are proposed based on different attributes.The feature points are further filtered and the relationship between frame and frame transformation is obtained.A double-layer reprojection error method is proposed for the confusion between dynamic points and mismatched points.The homography matrix is used as a projection matrix to perform error projection on the classification points,and the separation of uncertain points is achieved according to different feature attributes.Experiments show that the algorithm has a certain effect on dynamic targets with different speeds,and reduces the impact of large-area moving objects on target detection.Secondly,in order to adapt to various spatial states,a non-planar dynamic environment feature classification algorithm is proposed based on the plane feature classification algorithm,and the environment state is divided into four modes for feature classification respectively,and the ORB-SLAM2 initialization module is integrated to implement dynamic environment map initialization.jobs.Experiments show that the algorithm can effectively suppress the impact of dynamic features on pose estimation in a dynamic environment,and quickly and accurately implement map initialization.Finally,texture constraint and region constraint algorithms are applied to the ORB-SLAM2 trace thread.When the system performs pose estimation on the camera and three-dimensional points,the feature classification work is completed in advance,and the feature point advance pose estimation and three-dimensional point construction are performed using different attributes.A dynamic environment 3D SLAM algorithm is proposed in combination with the classification of dynamic and non-planar dynamic environment features and the dynamic environment map initialization algorithm.Through experimental research,it is verified that the algorithm can accurately calculate the pose change of the camera and construct an accurate sparse three-dimensional map.The research work of this paper can provide theoretical basis and method for the application of SLAM technology in dynamic environment.It has important theoretical value and scientific significance.