Research And Application Of 3D Reconstruction Of Small-Scale Objects Based On Polarization Imaging

As a hot research direction in computer vision,3D reconstruction technology plays an important role in social production and life such as unmanned driving,medical diagnosis,archaeological excavation and other fields.Compared with other 3D reconstruction techniques,the method of using reflected light polarization information to recover 3D models of target objects can accomplish high-precision imaging tasks using low-cost equipment,and has received more and more attention from researchers.In this thesis,we propose a method to recover the 3D model of an object using the polarization information of diffuse reflected light from the relationship between the polarization characteristics of reflected light and surface normals.The method combines stereo vision reconstruction technology to obtain the initial point cloud of the object as a priori knowledge to correct the ambiguous normal azimuths and reconstruct the 3D model with realistic information.First,the principle of polarization 3D reconstruction is investigated in this thesis.From the laws of refraction and reflection of light and Fresnel's formula,the polarization characteristics of reflected light are analyzed;the geometric model of diffuse reflection polarization 3D reconstruction is established,the relationship between the polarization information of the target object and the surface normal is studied,and the problems of using diffuse reflection light for polarization 3D reconstruction are analyzed;the 3D reconstruction method based on gradient integration technique is explored;a set of experimental system of diffuse reflection polarization 3D reconstruction is built to conduct experiments on small-scale objects.The experiments are carried out on small-scale objects,and the problems arising from the experimental results are analyzed.Secondly,to address the problem of ambiguity of the normal azimuth angle calculated by using polarization information,this thesis proposes a stereo vision normal correction method.The method uses stereo vision method to obtain the initial point cloud of the target object,calculates the target object normal a priori information,and corrects the ambiguous normal azimuth angle.Experiments prove that the method has good correction effect,and an accurate 3D model can be reconstructed with the corrected normal azimuth angles.Then,since the 3D reconstruction method based on gradient integration technique can only recover the relative depth of the target object,this thesis proposes a method to convert the relative depth into absolute depth by combining the initial point cloud of the target object obtained by stereo vision technique,and conducts an experimental analysis.In the experiment,the relative error between the calculated data of the 3D model and the measured data of the actual object is less than 10%,which proves that the method can recover the real information of the target object with high accuracy.Finally,based on the diffuse reflection polarization 3D reconstruction technology and the principle of stereo vision normal correction,combined with the actual requirements,this thesis designs and implements an application system for 3D reconstruction of target objects using polarization information.Users can use the system to obtain the image of the target object,adjust the parameters of the polarization stereo reconstruction algorithm according to the specific object,and view the instant reconstruction results through the 3D model display function.