The Research Of Key Issues On3D Reconstruction Based On The RGBD Images

The three-dimensional (3D) reconstruction technology mainly is referred to restore the3D information of the target objects from the image or the image sequence, which get moreand more widely used in computer vision, especially in computer animation, virtual reality,medical image analysis and other fields. However, the previous method which based on thetwo-dimensional images of visible light frequently encounter the influence of illuminationchange, shadow, object shelter and climate change et al, and cannot accurately identifythree-dimensional objects and its position, the depth information which calculated by texturecharacteristics, the variation of texture gradient, the integrity and the fuzzy of object canovercome above difficulties very well. Thus, how to obtain the accurate3D models quicklybecomes a research hotspot in recent years. In this paper, using the depth image and the colorimage (Hereinafter referred to as RGBD image) which obtained by the Kinect camera ofMicrosoft, aiming at the main question in reconstruction process, such as the pre-processing,the acquisition of point cloud data, the generation of the normal map, we propose a simplemethod for three-dimensional reconstruction.The main work is as follows:In this paper, we summarize the research background and related research status on3Dreconstruction at home and abroad, introduce the data obtained from the Kinect cameral, andexplore the reason for the noise generated of the data. On this basis, we use the bilateralfiltering method for noise reduction treatment, and filled in the holes of the depth data. In theprocess of calibration of depth data, we take the color image as the original image withoutdistortion, and proposing an calibration method based on the sparse feature points, and thusobtaining a mapping relationship between color images and depth images. Based on this, wecalibrate the processed data. Finally, according to the processed depth image, we get pointcloud data, and make the point cloud data to divide into triangulation, combine with the leastsquares fitting planar generating method to generate vector diagram, etc.By the implementing of the program and the analysis of results, this research of thispaper can make researchers reconstruct the real objects without expensive equipment, whatthey need are just Kinect and PC, to quickly realize the3D reconstruction, and effectivelyimprove the efficiency of3D reconstruction. As the results shown, although the3D model is not accurate enough because of lower accuracy, the general outline is clear enough to meetthe need of the usual research and daily application.