Research Of 3D Reconstruction Technology Based On Kinect Camera

3D reconstruction based-vision technology is widely used in artificial intelligence,unmanned driving,virtual reality,SLAM,and 3D printing,etc.It is one of the important directions for future scientific and technological development.Since Microsoft introduced the Kinect sensor,3D reconstruction technology based on RGB-D cameras has been widely studied and applied.This paper focuses on the research and improvement of 3D reconstruction technology based on Kinect V2 camera.The paper first introduces the hardware structure and working principle of the Kinect camera,then analyzes the imaging principle of the camera and derives its mathematical expression model,and then calibrates the two lenses of the camera respectively.The quality of the point cloud data depends on the quality of the depth map.Due to the camera structure,lens characteristics,or light in the environment,the depth image collected by the Kinect camera often has a lot of noise and holes.Therefore,this paper researches several currently used filtering algorithms.After simulation and comparison,it proves that using bilateral filtering can achieve good filling results.In the process of generating the point cloud,it is inevitable to generate some noise points and invalid discrete points.Therefore,only after dealing with the noise reduction and data compression of the point cloud data can the point cloud registration and visualization be better completed.In this paper,downsampling and removal of outliers are used to reduce the number of point clouds and effectively remove discrete points in the point cloud.Due to the high performance of the Kinect V2 camera,the amount of point cloud data collected is very rich,but when such a large number of point clouds are directly subjected to ICP registration,the amount of calculation is very large,which brings great challenges to the subsequent reconstruction work,so this article Improved for this situation.First,to solve the problem that the ICP algorithm requires high initial values of iteration,a rough registration algorithm of SAC-IA is proposed,and an initial rotation matrix and a translation matrix are first calculated.Secondly,the number and accuracy of registration points in the ICP algorithm directly affect the speed of iteration and the accuracy of the results.Therefore,this paper extracts and matches the feature points in the RGB image corresponding to the point cloud frame,and calculates the transformation relationship between the two frames.The mapping function between two-dimensional points and three-dimensional points is obtained,so as to find the corresponding points in the three-dimensional point cloud in the RGB image.When the two-dimensional points in a frame have no depth information,the point can be established by the Pn P algorithm.3D-2D projection model,calculate the3 D coordinates of 2D points.Then ICP registration is performed through these points,and the final registration model is obtained.Experimental results show that the speed and accuracy of reconstruction are significantly improved by the improved algorithm in this paper.For consumer-grade ordinary RGB-D cameras,the method of this article can still be used to complete the3 D reconstruction of indoor scenes to meet the needs of daily applications.