Research On 3D Reconstruction Of Non-rigid Scenes

With the rapid development of computer vision technology and computer graphics technology,3D reconstruction technology has been widely used in many fields.3D reconstruction is generally divided into 3D reconstruction of static rigid scenes and 3D reconstruction of dynamic non-rigid scenes.Compared to the reconstruction of rigid scenes,the reconstruction of non-rigid scenes can recover 3D information of moving objects,which has a larger research significance and practical value.The existing non-rigid 3D reconstruction algorithms have the problem of inaccurate registration during the non-rigid registration process.At the same time,distortion will occur during the model surface fusion process.The 3D model needs to be optimized based on the illumination information of the surface of the object.It is of great significance to study the non-rigid registration process and surface illumination estimation method to improve the quality of non-rigid reconstruction.This thesis focuses on the factors that improve the 3D reconstruction effect of non-rigid scenes,proposes a non-rigid registration method based on near-infrared information and a surface illumination estimation method based on near-infrared information,improves the performance of non-rigid registration,and provides a more accurate basis for model surface optimization.Aiming at the problem that the regular term of the non-rigid registration process is too strong.The thesis analyzes the method of constructing the energy function for the non-rigid deformation domain in the non-rigid registration process.An optimization scheme based on near-infrared information is proposed to solve the problem of excessive smoothness and loss of details of the reconstructed model due to the strong constraint of the regular term in the energy function,adjust the constraint intensity of the regular term according to the change of surface near-infrared light intensity.Then,the solution process of the optimized energy function is studied.Finally,a non-rigid 3D reconstruction system is established to test the proposed optimization algorithm performance from the average numerical error of the model,the single frame processing time,and the model reconstruction quality.The experimental results show that the algorithm proposed in this thesis reduces the average numerical error of the model by 0.0884 cm on average,compared to the Dynamic Fusion algorithm.There is no significant change in the time consumption of the single frame processing.The quality has also been significantly improved.Aiming at the problem of insufficient consideration of nonlinear response when establishing an illumination model.First,the imaging stability of near-infrared images and the reflection characteristics of near-infrared light on the surface of the object are analyzed.Then,the illumination characteristics of near-infrared light attenuation with distance are estimated.Based on the conditions of the infrared light source and the physical propagation process of near-infrared rays in space,estimate the lighting model.The experimental data is collected to fit the non-linear response parameters and determine the mathematical expressions of the lighting model were given.Finally,the scene surface reflectance was classified using the lighting model.Experimental results have demonstrated that compared with the algorithm of Choe et al.,the illumination model established in this paper can achieve an increase of PLCC by 0.0099 and SROCC by 0.0038.At the same time,the classification effect has also been improved to a certain extent,which indirectly reflects the better performance of the illumination model in this thesis.