| Polarization three-dimensional(3D)imaging is an important part of 3D information acquisition,it is well developed and broadly used in many areas such as industrial quality inspection,aviation terrain mapping,face recognition,and security monitoring.Because it is ambiguous when calculates the normal vector of the object surface,which is caused by the ambiguity when calculating the arctangent.The reconstructed 3D result is distorted.That’s why the ambiguity should be limited.And how to solve the ambiguity of the normal vector in different complex circumstances,as well as improving the quality of the reconstruction is meaningful research in this area.This thesis studied the key technology of polarization 3D imaging of complex scenes to solve the 3D imaging problem when the object with an uneven albedo surface or complex microstructure,meanwhile,achieve the 3D imaging when multi-object existed.We established a model that described the variation of polarization characteristic of the reflected light from different types of surface.The relationship between the different polarization characteristics of the reflected light and the normal vector of the object surface was particularly learned.Based on this,the near-infrared polarization 3D imaging model was proposed combined with the stable variation trend of the light intensity reflected by the uneven albedo object.Then the imaging problem for the complex micro-structure object was solved by using the polarization 3D imaging model with prior information about the low-resolution depth.When multi-object existed,the 3D imaging was achieved by using the relation between the imaging distance and the degree of blurred edge.The main work and innovations in this thesis include.(1)Starting from the physical mechanism of the interaction between a light wave and different types of surface,the transmission of light on smooth object surfaces and rough object surfaces was analyzed,and a 3D imaging model based on the polarization characteristic of the diffuse reflected light was established to verify the polarization 3D imaging for diffused object surface.It provides a theoretical basis for using the polarization characteristics of the reflected light to achieve 3D contour reconstruction.(2)The relationship between various color regions and the intensity of reflected light under different light wavelength bands was analyzed for the object with different colors.It showed that the reflected light intensity of different color areas in the near-infrared band is relatively stable.Therefore,a near-infrared polarization 3D imaging model was proposed.This method calculated the normalization factors according to the stability of intensity from the different areas.Then the factors were introduced into the reconstruction process of normal vector calculating,which solved the polarization 3D imaging problem for colorful objects.Besides,the accuracy of the reconstruction was analyzed,and we figured out the relation between the accuracy and the imaging distance.A colorful object polarization 3D imaging experiment was conducted,and the experiment results showed the reliability of the former content.(3)The polarization 3D imaging method based on prior information of the gradient field of the low-resolution object of the deep network was proposed for the complex micro-structure object,which avoided the problem that the intensity disturbed each other,and the gradient referenced information was hard to obtain accurately.This method limited the ambiguity of the polarized normal vectors by using the low-resolution gradient information from the deep learning,therefore,achieved the high accuracy reconstruction of the surface with complex micro-structure.(4)We analyzed the relationship between different object distances and the blurred degree of the edge when multi-object existed.According to the knowledge of the geometrical optics and deep learning,the estimating of the imaging distance,as well as the deblurred images,was obtained by using the degree of blurred edge.Based on the accurate acquisition of the imaging distance and reflection characteristics,a polarization 3D imaging method using the blurred degree of the edge was proposed for scenes that multi-object existed.The experiment showed the availability of different object distance recovery and the 3D contour reconstruction for the multi-object scenes. |
