Study On The Key Technologies Of Polarization 3D Imaging Using Diffuse Reflection

The three-dimensional reconstruction technology which refers to recover the shape of an object has always been an important research field in computer vision.It has a wide range of application prospects.Nowadays,the technologies which make use of image shadows,textures,polarization and motion information to carry out three-dimensional reconstruction are the main trend for the shape recovery methods based on characrastics of images.Meanwhile,shape from polarization has been widely researched with the use of polarization in shape recovery.The shape from specular polarization is the mostly used method in three-dimensional reconstruction recently.However,it is hard to realize the shape recovery from specular reflection because the specular reflection strongly relies on the direction of the detection where the incident light must be strictly controlled,making this method quite complicated to achieve.In this thesis,a method called shape from diffuse polarization is proposed to recover the shape of objects by using the diffuse reflection component of the reflection from objects,and the depth information with low resolution acquired by Kinect is used as the prior knowledge to correct the normal map obtained from shape from diffuse polarization.A remarkable result of reconstruction is obtained after the correction and the post-processing.The main work of this thesis is as follows:(1)The theoretical basis of shape from diffuse polarization is discussed.The polarization state of light waves and its representation methods are analyzed.The derivation process of Stokes vector representation method of polarized light and its measurement method are given,and its relationship with geometry representation of polarized light is explained.The feasibility of shape from diffuse polarization is proved theoretically by using the Fresnel formula to solve the refractive index and reflective index.(2)The geometrical model about the relationship between the polarization state of the reflection light from objects and their surface normal is established.By using the Fresnel formula,the relationship between the degree of polarization from the specular reflection light or the diffuse reflected light with incident angle is built.The errors of the surface normal of the object obtained by using the degree of polarization of the specular and diffuse reflected light are discussed,and the method using Kinect depth information as the prior knowledge to calibrate the normal map is put forward to solve the problem.(3)The experimental platform is built based on the theory of shape from diffuse polarization in this thesis,meanwhile,the reconstruction process and experimental method of shape from diffuse polarization are given.The existed problem on the recovered shape obtained by reconstructing the uncorrected normal map is analyzed.Then Kinect depth information is utilized to correct the normal map,and the corrected shape from reconstruction is obtained.The effect of noise and specular reflection on the experiment results is analyzed.Finally the results of shape recovery are improved by image denoise and separating the specular reflection component from the image.The validity of shape from diffuse polarization and the reasonability to correct the normal map with Kinect depth information are verified.In this thesis,the three-dimensional reconstruction of the surface of the non-transparent insulating object is realized by using shape from diffuse polarization method.The ambiguity of the incident azimuth angle in this method is solved by the fusion of deep information from Kinect.The improvement method is presented by studying the effects of noise and the specular component of reflection.The experimental condition for the method proposed in this thesis is simple and the result is remarkable,which provides a solution to the problem of the loss of detail in low resolution depth sensors.