Research On 3D Reconstruction Algorithm Based On Depth Image

Depth data is the embodiment of length and distance in spatial coordinates,and 3D reconstruction based on depth information in depth image is to reproduce the surface and contour information of the object using the depth data information contained in it.It has a prospective and economic value that can not be underestimated in medical bone repair of non-contact organs,virtual experience in daily life and entertainment,and reconstruction of 3D printed objects in scientific research.In the increasingly developed modern life of intelligent products,in the field of computer vision and human-computer interaction,which is close to life but related to everyone's own interests,the face recognition technology involving in-depth information is the hot spot in the research,but it is also a challenge to improve the accuracy.This paper focuses on the improvement of 3D scene surface reconstruction based on depth image information,including the preparation of 3D reconstruction,camera calibration,preprocessing of collected depth image,point cloud registration and fusion after depth data is converted into point cloud data,3D object surface generation,etc The following was done:In this paper,we use zed camera to collect depth image information.For the important part of camera calibration,according to the characteristics of Zhang Zhengyou camera calibration method,we introduce image interpolation algorithm to correct the distorted image information in the nonlinear camera calibration.A mean filter depth image restoration algorithm based on pulse coupled neural network is proposed for the hole area in the collected depth image information.The hole area is identified first,and then the depth image is repaired by adaptive mean filter.The depth image information collected and repaired is transformed into point cloud information,and rough registration of point cloud is carried out by combining normal evaluation with histogram feature point extraction and RANSAC algorithm.The rough range of target point cloud is selected first,and then point cloud fine registration and point cloud fusion are carried out by ICP algorithm,so as to get more accurate point cloud collected from multiple perspectives.For the results of point cloud fusion,this paper applies K-means clustering algorithm to the point cloud processing,identifies outliers and then removes them,providing more perfect point cloud information for 3D reconstruction.So far,we have obtained the point cloud data from multiple perspectives,but this result is not close to people's daily habits when using.Therefore,after comparing the commonly used object surface reconstruction algorithms,we choose the greedy projection triangulation surface reconstruction which is suitable for the characteristics of this paper to complete the triangulation of point cloud data,which completes the whole flow of 3D reconstruction based on depth image Cheng,we can get a realistic three-dimensional surface model of the object.