Fast 3D Texture Reconstruction Based On A Single Depth Sensor

The research of 3D reconstruction has always been an interesting topic and it is also a difficult and hot spot of computer vision recently.This technology can provide us with vivid models and has a promising future.In this paper,a 3D reconstruction algorithm based on a single depth sensor is studied.This Paper uses Kinect as the depth device and gets the depth data and RGB data with the objects that will be reconstructed.We can get the 3D models after processing the raw data with filtering,registration,fusion and RayCasting.The main work of this paper is as follows: the method of 3D reconstruction is based on the algorithm of kinectFusion.This algorithm divides 3D reconstruction into several steps,such as acquisition and transformation of raw depth data,camera tracking,point cloud integration,raycasting and marching cubes.Firstly,we get the depth data and RGB data of objects with Kinect.Secondly,we filter the noise using bilateral filtering,then we convert the depth maps into vertex maps and normal maps which are convenient to be processed.Thirdly,we register the data using iterative closest points,then we integrate the data,which is used to convert different frames into the same coordinate system,and then fusing all the frames to get a complete 3D model.This paper adds texture information to the 3D model,which makes the model vividly.The algorithm of kinectFusion has accumulated error which leads to closure loop.To solve this problem,this paper uses a method based on the camera trajectory and spits the trajectory into segments to distribute the residual.