Stereo Vision

Stereo vision studies how to infer depth and 3D structure by matching corresponding features across calibrated viewpoints, making it useful for robotics, mapping, medical imaging, and measurement. Methods range from classical epipolar geometry and triangulation to deep stereo networks, attention-based feature fusion, event-camera systems, and learned models that use synthetic or monocular data. Researchers evaluate these approaches on benchmarks such as KITTI, Middlebury, and ETH3D, as well as in settings including underwater imaging, surgery, and robot manipulation. A central challenge is balancing accurate, reliable geometry under occlusion, changing illumination, and domain shift with the speed, calibration quality, and data efficiency needed for real-world use.