Research On Indoor Scene Reconstruction Technology Based On Kinect

With the movement and popularity of consumer cameras,the introduction of 3D technologies is making it easier and cheaper and 3D reconstruction technology has gradually been applied to many social fields such as medical skeleton models,cultural relics and building restoration,game scene design,online and offline real estate marketing,etc.However,when the Kinect camera obtains depth data,it is limited by factors such as hardware,ambient lighting conditions,and distance,and the depth data obtained is often noisy and empty,and the geometric depth information is seriously missing,which brings difficulties and inaccuracies to indoor 3D modeling.At present,the commonly used indoor scene 3D reconstruction methods have problems such as cumulative error,wrong matching,poor dynamic real-time effect,limited scene reconstruction size,no color texture information and slow data processing speed.Therefore,it is of great theoretical significance and practical value to conduct in-depth research on the 3D reconstruction technology of indoor scene based on Kinect depth camera.This thesis mainly focuses on the related problems in indoor 3D reconstruction based on Kinect depth camera.In order to ensure the correctness of the output data of the depth camera,the correction and registration of the camera lens distortion is completed on the basis of the camera imaging model.An image restoration algorithm based on color and depth information combined with guide map and joint bilateral filtering is designed to remove noise and fill the voids in the collected depth images.Aiming at the problem of low matching accuracy of point cloud data,RANSAC is used to eliminate abnormal points and provide good initial values for fine registration.Aiming at the problems of large cumulative error and slow point cloud fusion,an improved point cloud registration algorithm combining SICP and Levenberg-Marquardt is proposed to achieve accurate matching of point cloud data,improve registration efficiency and reduce registration error,and then combine with TSDF model to realize three-dimensional reconstruction point cloud registration and fusion rendering.In order to verify the effectiveness of the proposed 3D reconstruction algorithm,the real-time verification of the algorithm was verified by scanning and reconstruction of the real indoor scene,and the reconstruction performance was compared with the mainstream Bundle Fusion algorithm on the TUM public dataset.The experimental results show that the algorithm proposed in this paper has smaller relative pose error and absolute pose error,the generated model is clearer and more accurate,and it is more suitable for three-dimensional reconstruction in indoor scenes,and the algorithm has good robustness and real-time.