Research On Real-time Construction Algorithm Of Indoor 3d Dense Point Cloud Based On V-SLAM

With the complexity of the application environment,human beings put forward higher requirements for the autonomy of mobile robots,and hope that autonomous mobile robots can perform tasks in the initially unknown complex environment,such as safe and effective interaction with humans,other vehicles and their environment.Based on this demand,the integration direction of 3D reconstruction and visual SLAM has gradually become one of the hotspots and trends studied and discussed by scholars.In order to realize the indoor 3D dense point cloud real-time construction and path planning of mobile robot,the following work is carried out for Kinect 2 camera and SLAM technology:Firstly,this paper calibrates the internal and external parameters of Kinect 2camera.Because Kinect 2 camera will deviate from the depth value calculated by the camera projection model when measuring the depth,a linear model d=ad+βis proposed to correct the measured depth value,so that the measured depth value of Kinect 2 camera is consistent with the depth value calculated by the visual odometer,More accurate depth information is obtained to reconstruct the 3D dense point cloud of the later indoor environment.Secondly,in the front-end visual odometer,aiming at the problem that ORB feature points mainly focus on rich texture when extracting features,the quadtree strategy of uniform extraction of ORB features is studied in detail;In terms of feature matching accuracy,GMS feature matching algorithm is proposed to eliminate mismatched feature point pairs.Through the comparison with RANSAC feature matching algorithm,the matching accuracy and polar constraint are used to verify,in which the matching accuracy is 89.92%and 94.21%respectively,and the polar constraint is 0.001315 and 3.33e-5 respectively.The results show the effectiveness and feasibility of the algorithm.Then,aiming at the problem that the 3D sparse point cloud can not be applied in the later robot real-time navigation,this paper designs an indoor 3D dense point cloud real-time construction algorithm framework based on Kinect 2 camera.In the aspect of camera pose estimation,Pn P is used to estimate the local camera pose;In the key frame filtering of SLAM,the filtering condition based on distance judgment is adopted to improve the mapping efficiency of the system;In the back-end optimization,Pose Graph optimization is used to optimize and estimate the global camera pose;In order to ensure the consistency of global trajectory,in loop detection,Bo W3(Bag of Words)model is used to judge the scene similarity,and the similarity score is improved.Finally,the proposed algorithm is verified by experiments.Comparative experiments are carried out on different data sequences in the TUM dataset.In the experimental results of this algorithm,the absolute trajectory error is less than the absolute trajectory error of ORB-SLAM2,the RMSE is improved by 17.4%on average,and the three-dimensional dense point cloud is reconstructed for the TUM dataset and the real scene in the laboratory,The results show that the SLAM system based on RGB-D vision sensor can meet the needs of mapping and positioning.Then,in order to reduce the calculation cost of mobile robot’s later path planning,the point cloud map is transformed into octree map,and then transformed into two-dimensional grid map through the oblique projection principle.The indoor path planning of mobile robot laboratory is realized by using A*algorithm.