Global Registration Cumulative Error Minimization Based 3D Human Body Reconstruction Using RGB-D Scanning Data

3D human body reconstruction using RGB-D scanning data is still challenge due to low-quality data,noise interference and registration error.To solve these problems,a global registration cumulative error minimization based 3D human body reconstruction method using RGB-D scanning data is introduced.The first step is to capture and preprocess human body scanning data,eliminating noise and background.Then a method based on point feature descriptor is proposed for local rough registration and this pairwise registration is refined by iterative closest point approach.Afterwards,a weighted graph for local aligned data is constructed and local data in this graph is combined using minimal spanning tree.This reduces the influence of global error propagation.A global rigid registration is achieved based on loop closure.In order to solve body drift problem in scanning and further reduce cumulative error,an incremental non-rigid method is applied to global non-rigid registration.Finally,3D human body model is generated based on global aligned data and color information in scanning data.Using two Kinect scanning devices to cover full body,the proposed method can conveniently and effectively reconstruct highly realistic 3D human body and the biometric measurements error between reconstructed bodies and real bodies is small.