Research And Implementation Of High Quality 3D Texture Reconstruction

3D reconstruction technique is widely used in a great number of applications such as AR and SLAM.In recent years,the advent of the Microsoft Kinect and other consumer grade RGB-D cameras has resulted in tremendous stride in the field of 3D reconstruction.By using the algorithm like the Kinect Fusion,people can create geometric model of the real-world object and scene with high-fidelity.The 3D reconstruction process consists of geometric reconstruction and texture reconstruction.For the producing of colored geometric models,most existing system use the volumetric blending approach which integrate the color information of each pixel into a voxel like the way of depth integration in KinectFusion.With the advantage of volumetric representation,this approach can produces the colored 3D object in real-time,but the result model often suffer from motion blurring,ghosting and other artifacts.In this paper,we describe an optimized texture reconstruction approach.Our approach concentrates on the difficulties in the volumetric blending approach.To resolve the motion blur in the input color image,we add blur detection as a preprocess module to select the color frames which are not suffer from motion blur.To resolve the ghost on the color model,we correct the updating error on the color image by using a new weight function for the color in the voxel.We have evaluated our approach between the original approach on different scene and object.The resulting models show that our approach contains a high-quality color map with few blur and ghost.