Xuan Li

Shape from polarization (SfP) inherently encounters surface normal azimuth ambiguity, which hinders accurate depth estimation. Traditional methods attempt to resolve the ambiguity by incorporating prior information through high-constraint pixel-wise normal correction, making their accuracy heavily dependent on the quality of the prior. In this paper, a low-constraint prior-guided robust polarization 3D imaging framework is proposed to address ambiguity by combining gradient analysis with geometric constraints. Unlike the traditional method, our method performs regional processing to achieve accurate normal estimation, which significantly decreases sensitivity to priors, thus allowing for application to objects with a specular-diffuse hybrid model. Several experiments were conducted on both synthetic and real datasets to evaluate the effectiveness of our method, and the results indicate remarkable performance, achieving an approximately 20-fold reduction in dependence on prior information compared to the traditional method.

Polarization is closely correlated with surface morphology, so polarization information has been increasingly integrated into stereo imaging technologies. Nonetheless, the lack of a systematic, unified, and intuitive framework for assessing the accuracy of polarization-based stereo imaging hinders a reliable quantitative evaluation of imaging performance and limits performance assurance and controllability in future engineering applications. This paper proposes a quantitative evaluation method for polarization-based 3D imaging accuracy that takes both imaging device conditions and imaging quality into consideration. The validity and robustness of the proposed method are verified through simulations and experiments. Our method provides a solid foundation for the design, optimization, and engineering deployment of polarization stereo imaging systems, facilitating the transition of polarization-based 3D imaging from laboratory research to controllable engineering applications.

Polarization-based 3D reconstruction offers high precision and low cost, yet existing methods overlook noise suppression during image acquisition. We propose a pixel-wise multi-exposure fusion method that selects optimal responses within the linear camera curve to reduce photon noise and nonlinear distortions. A highlight correction strategy further mitigates specular-gradient errors. Experiments show that the method effectively suppresses Gaussian and systematic noise, enhances polarization accuracy, and yields more precise surface normals and clearer 3D reconstruction details. Specifically, the root mean square error (RMSE) of the reconstructed surface normals is reduced by approximately 14.3% relative to the traditional method, which quantitatively demonstrates that our reconstruction results are closer to the ground truth.

Multi-view stereo (MVS) estimates depth by matching features across calibrated views. Though highly accurate on textured surfaces, it is prone to holes and coarse geometry in the case of weak textures or limited viewpoints. Shape-from-polarization (SfP) recovers dense normals from polarized reflections regardless of texture but causes ambiguities in angle inversion, demanding additional disambiguation. Based on these complementary strengths, this paper proposes a passive 3D reconstruction framework that fuses coarse but globally consistent depth priors from a self-attention-enhanced PatchMatch network with fine-grained normal gradients recovered from calibrated polarization measurements. These cues are incorporated into a joint optimization that improves linear consistency between depth and polarization-derived normals while applying a robust, graph-based spatial smoothness constraint to address azimuthal ambiguities and suppress outliers. The final surface is acquired by integrating the optimized normal gradient field in the Fourier domain. Experimental results on different weakly textured objects indicate that our method obtains finer details and fewer artifacts than advanced multi-view stereo and deep learning-based methods, with significantly fewer viewpoints and less computational overhead.

We propose a Monte Carlo-based method for recovering polarization information in scattering scenes. This approach introduces a backscattering model to characterize polarization state variations within scattering media. Specifically, the Monte Carlo algorithm is employed to construct a computational model for simulating and analyzing the backscattering process of incident light within the scattering medium. By leveraging the mathematical relationship between the Stokes vector and the Mueller matrix, we mitigate the effects of backscattering on polarization information. The efficacy of the proposed methodology is validated through experiments. The findings demonstrate that the proposed method successfully recovers target polarization information while achieving enhanced imaging contrast compared to conventional polarization-difference imaging techniques.

The structured light 3D measurement technology based on laser galvanometer has been widely used in industrial scenarios such as robot grasping, handling, and loading/unloading in welding and assembly workshops. However, in actual measurement scenarios, there are complex measurement structures such as depressions, overlaps, and occlusions. Light is prone to multiple reflections between micro-faces, causing intensity information to be mixed within the micro-face area and ultimately resulting in point cloud loss in this measurement area. To address the issue of point cloud loss in complex structure areas in the measurement process and ensure the accuracy of the measurement information provided by vision, a binocular point cloud sensor with a laser galvanometer as the key projection module is proposed in this work. Without adding hardware, it realizes two different image projection modes to deal with complex measurement situations within the scene. Among them, the anti-multiple reflection projection mode proposed in this work, by regulating the timing coordination relationship between key components, completes the measurement of complex structure positions and solves the problem of point cloud loss caused by multiple reflection interference. Finally, multiple experiments are conducted in actual scenarios to verify the feasibility of the proposed strategy. The experimental results show that in measurement scenarios with multiple reflection interference, the integrity of the black part point cloud measured by the anti-multiple reflection projection mode proposed in this work reaches 98.03%, which is 18.98% higher than the traditional measurement mode. It effectively solves the problem of point cloud loss in measurement scenarios with multiple reflection interference. Visual measurements ensure the accuracy and completeness of the information obtained. The six-axis compensation values determined by the robot for each part’s pose state during teaching become more precise. This ensures that the previously taught robot trajectory can be accurately reused for subsequent poses, thereby reducing the time needed for manual robot debugging and enhancing production efficiency.

Photonic lantern is a key device in space division multiplexing (SDM) system. The key challenge of a photonic lantern is mode scalability, which requires the taper length to increase nonlinearly as the mode number scales up. The traditional photonic lantern fabrication method requires stacking the input fibers into the hollow, low-index outer cladding before tapering. It implicitly sets geometric constraints on the input fibers’ core positioning. We propose a photonic lantern design with drilling preform and reduced cladding fibers to lift these constraints and make photonic lanterns more adiabatic. By analyzing the effects of loosening the constraints on the adiabatic requirement of a three-mode photonic lantern, we find further progress could be made to alleviate this adiabatic requirement. The optimal structure for our design is proposed and demonstrated through the beam propagation method (BPM). Our findings could help further improve the mode scalability of photonic lanterns.

Polarization 3D imaging is a passive, monocular, long-distance 3D imaging technology. Compared with traditional 3D imaging methods, it has many advantages, such as its lack of need for a light source, lack of need for image matching, and ability to achieve 3D imaging using only a single image. In this study, the principle of polarization 3D imaging was introduced. In the design process of a polarization 3D imager, the acquisition method for obtaining polarization information, the extinction ratio, the spatial resolution, and the refractive index of objects was introduced in detail. The influence of these key factors on the accuracy of polarization 3D imaging was analyzed. Taking the limitations of a small satellite payload into account, specific indicators such as multi-aperture polarized imaging, a 10,000:1 extinction ratio, and a spatial resolution of 30 m were designed. The implementation and functions of the polarization 3D imager were elaborated upon, and optical systems and polarizing devices were developed. Finally, by utilizing the image data obtained by the polarization 3D imager, polarization 3D imaging of real ground objects was obtained. The accuracy of the polarization 3D imaging inversion was approximately twice the spatial resolution. These research results lay the technical foundations for the development and practical application of polarization 3D imaging technology and instruments.

Polarization 3D imaging has been a research hotspot in the field of 3D facial reconstruction because of its biosafety, high efficiency, and simplicity. However, the application of this technology is limited by the multi-valued problem of the azimuth angle of the normal vector. Currently, the most common method to overcome this limitation is to introduce additional depth techniques at the cost of reducing its applicability. This study presents a passive 3D polarization facial imaging method that does not require additional depth-capturing devices. It addresses the issue of azimuth ambiguity based on prior information about the target image's features. Specifically, by statistically analyzing the probability distribution of real azimuth angles, it is found that their quadrant distribution is closely related to the positions of facial feature points. Therefore, through facial feature detection, the polarized normal azimuth angle of each pixel can be accurately assigned to the corresponding quadrant, thus determining a precise unique normal vector and achieving accurate 3D facial reconstruction. Finally, our azimuth angle correction method was validated by simulated polarization imaging results, and it achieved accurate correction for over 75% of the global pixels without using additional depth techniques. Experimental results further indicate that this method can achieve polarization 3D facial imaging under natural conditions without extra depth devices, and the 3D results preserve edge details and texture information.