Real-time High-accuracy 3D Reconstruction Based on Texture and Geometric Features

We present a robust approach for reconstructing accurate 3D model based on image features combine with geometric features using RGB-D camera.First,the depthimagecapturedby RGB-D camera is filtered by our efficient algorithm to remove the intrinsic noise. This allows us to perform high-fidelity reconstruction. Then the point cloud is extracted texture and geometric features from color and depth image. Meanwhile, some mismatch point pairs are filtered by point color and distance statistics information. A new point cloud registration based on internal relative vector is proposed to obtain the current camera pose. In additional, a voxel count algorithm is proposed to register feature point and acquire an excellent fusion result. With the registered feature point, the 6DoF pose can be easily estimated from RGB-D image. Experimental results demonstrate that the proposed method have an excellent in both reconstruction and pose estimation.