Research On 3D Reconstruction Technology Based On RGB-D Camera

With the improvement of hardware performance and the maturity of related fields,the 3D reconstruction technology based on RGB-D cameras has become the key to technological innovation in the field of 3D reconstruction with its high precision and dense reconstruction.In proposed 3D reconstruction systems based on RGB-D cameras,the quality of the input depth map and the accuracy and efficiency of the deformation domain estimation are the key parts that directly determine the system performance.However,the depth maps obtained by the RGB-D cameras generally have problems such as low resolution,noise,and lack of depth values,which cannot provide rich and accurate spatial structure information.At the same time,for the non-rigid scenarios in the real world,the scale of the deformation domain estimation problem is much larger than the scale of the rigid deformation domain estimation problem,meanwhile,since the efficient solution for the state estimation of the non-rigid deformation domain has high commercial value,the design of this part is not disclosed in previous schemes,which forms a technical barrier and greatly hinders the development of non-rigid 3D reconstruction.In order to improve the system performance,this paper proposes a NIR image-guided depth map enhancement algorithm for a 3D reconstruction system based on RGB-D cameras and proposes an efficient solution for state estimation in the non-rigid deformation domain.In order to provide more accurate scene spatial structure information and improve system performance,this paper proposes a NIR image-guided depth map enhancement algorithm combining near-infrared characteristics and the idea of autoregressive models.The algorithm is based on an autoregressive model,using the reflectance characteristics of near-infrared light and the robustness of near-infrared light in an environment with suboptimal ambient light,combining the shadow detection of dark light photography to construct an optimization model that matches the characteristics of near-infrared light to solve.The experimental results show that the method proposed in this paper can avoid the gradient information that is inconsistent with the depth map in the previous algorithms based on RGB image guidance,suppress the occurrence of "artifacts" in the process of enhancing the depth map effectively,and at the same time,without introducing wrong edge information,the depth map enhanced by NIR image guidance also retains the edge information of the original depth map better.In a non-rigid 3D reconstruction system,the state estimation of the non-rigid deformation domain is directly related to performance of the system.In order to ensure that the system can reconstruct objects accurately and efficiently,this paper proposes an efficient solution for state estimation in the non-rigid deformation domain.This scheme derives the corresponding non-rigid deformation domain construction formulas to obtain the corresponding Jacobian and Hessian matrices.An analytical formula based on the Levenberg-Marquardt optimization method is constructed and a solution algorithm is constructed based on the pre-processing conjugate gradient algorithm.Based on the above work,a parallel solution architecture for non-rigid deformation domain is designed through CUDA.The experimental results show that the efficient solution based on parallel computing proposed in this paper can effectively reduce the time consumed by the non-rigid deformation domain estimation process while ensuring accuracy,and is basically not affected by the size of the problem,ensuring the system's efficiency and robustness.