For objects with smooth surfaces, traditional reconstruction algorithms cannot accurately restore their shape characteristics, and the use of polarization information to reconstruct the target object will solve the above situation well. Due to the ambiguity of the azimuth angle of the incident surface, binocular estimation is proposed. Parameter removal of ambiguity angle, the Stokes parameter is used to represent the polarization state information under two viewing angles. The relative pose is estimated from two images with polarization information. The relative rotation and translation are indispensable, then the problem can be converted. For the least squares optimization problem, the azimuth angle ambiguity problem is eliminated by obtaining the optimal solution. The experimental results show that the image resolution of the depth map is higher after the ambiguity is eliminated by the algorithm, the shape information after reconstruction is accurate, and the target object texture. It is highly reducible and easy to realize in engineering.