3D Reconstruction Of Indoor Scene Based On RGB-D Camera

With the rapid development of computer technology and artificial intelligence,there is a growing demand for intelligence,especially for the perception and understanding of complex dynamic scenes.For relatively closed Spaces such as home life,offices,supermarkets and factory workshops,the 3d reconstruction of indoor scenes is an important application direction,and it is also the application basis for indoor robots to realize accurate navigation planning and scene object recognition in complex environments.However,the indoor complex environment is affected by factors such as shade,lighting and the camera itself.Therefore,it is an urgent technical difficulty to complete the indoor 3d reconstruction quickly and accurately.The specific work of this paper is as follows:Firstly,in view of the problem of indoor scene 3d information collection,this paper USES the Microsoft Kinect camera to scan indoor scene,so as to obtain the color and depth 2d image information of the scene.The internal parameter matrix of the Kinect camera is obtained through the checkerboard calibration method,and the depth map is converted into 3d point cloud data,so as to complete the data collection of indoor environment.Then,Then there are noise and hole problems for the depth image.However,the traditional algorithm fails to effectively use the global information and neighborhood relationship provided by the color image,and can not effectively segment the object boundary,which is easy to produce a smooth phenomenon,which affects the repair effect of the depth map.A problem,this paper uses a fuzzy C-means clustering segmentation guided depth image for repair.Firstly,by introducing a color image,the traditional anisotropic filter repair algorithm is improved,and the depth map is globally optimized to eliminate large-area hole regions.For discrete hole points,the color image is divided into multiple similar regions to obtain the structural information in the scene,and the traditional bilateral filter repair algorithm is reconstructed to control the weight image to guide the further enhancement of the depth image.The experimental results show that the algorithm can improve the operation speed under the premise of ensuring the filtering effect.Finally,aiming at the problems of slow speed and low precision of the current point cloud registration algorithm,this paper improves the iterative nearest registration algorithm,USES k-d tree to speed up the search of corresponding point pairs,and eliminates the mismatching points by combining the double constraints of curvature and point distance,so as to improve the registration accuracy.Through experimental verification,this paper combined the improved iterative nearest registration algorithm with the sample-based registration algorithm,greatly improving the registration accuracy and speed of the high point cloud,and thereby improving the effect of indoor scene 3d reconstruction.