Real-Time 3D Reconstruction Based On RGB-D Camera And Its Application In Robot

3D reconstruction can be used to obtain 3D model of objects.Real-time 3D reconstruction plays an important role in the field s of robotics,VR and AR.With the rapid development of these fields,more and more attention has been paid to real-time 3D reconstruction.Low-cost RGB-D cameras,like Kinect,can simultaneously acquire depth image and RGB image,making it ideal for real-time 3D reconstruction.Low-cost RGB-D camera greatly promoted the development of real-time 3D reconstruction,but many problems are not well resolved.Moreover,depth images obtained by RGB-D are not accurate enough.3D reconstruction requires high performance computer.3D reconstructed models have poor accuracy.We aim at real-time 3D reconstruction algorithm based on low-cost RGB-D camera.We designed a new real-time 3D reconstruction algorithm using a RGB-D camera.For an indoor environment,we use existing geometric planes to optimize real-time 3D reconstruction.We use truncated directional distance field(TSDF)to store 3D model.We use geometric planes for depth image optimization and initial registration,and then we use direct registration method for accurate registration.We use geometric planes to optimize accurate registration and data fusion.Geometric planes can be used for operations after scanning,such as segmentation and meshing.We use GPUs to further accelerate 3D reconstruction to achieve real-time performance.Compared to Kinfu,our algorithm can achieve better reconstruction results,especially for the scenes with more geometric planes.Finally,we apply our algorithm to a robot application,we can real-time remotely reconstruct 3D model.