Polarization-vector-guided stereo matching in textureless scenes

Disparity matching in stereo vision is often limited by feature information, especially when dealing with textureless scenes. To address this challenge, we propose a polarization-vector-guided stereo matching framework. By exploiting the strong correlation between the polarization properties of the reflected light field and the geometric features of object surfaces, a matching model based on minimizing the inner product of polarization normal vectors is established to solve stereo matching problems for weak-texture and textureless scenes. Simulation and experimental results demonstrate that the proposed method uses polarization information to reconstruct the surface normal vector field with high accuracy, thereby reducing matching failures in stereo vision without relying on any texture features or prior information of the scene.