Active Stereo

Active stereo vision aims to reconstruct 3D scenes by analyzing images from two cameras, often enhanced with projected patterns. Current research focuses on improving accuracy and robustness, particularly addressing challenges like occlusions and noise, through techniques like pseudo-stereo inputs, Markov random field optimization, and novel neural network architectures such as SwinIR-based models. These advancements are driving progress in applications ranging from autonomous vehicles and robotics to medical imaging, where accurate and reliable depth estimation is crucial. The development of physics-grounded simulators and mixed-domain learning approaches is also significantly improving the training and generalization capabilities of active stereo systems.