Stereovision

Stereovision is the perception of depth and three-dimensional structure produced by combining two slightly offset images, typically captured by the left and right eyes or by parallel cameras. In humans, this process relies on stereopsis, where the brain interprets binocular disparity—the difference in the position of corresponding points between the two images—to infer depth. In computer vision, stereovision denotes a set of techniques that compute depth from a rectified stereo image pair.

Core concepts include camera calibration, rectification, and disparity. Rectification aligns epipolar lines so corresponding points lie on the same horizontal scanline, reducing the search for matches to a one-dimensional problem. Disparity is the horizontal shift between corresponding pixels, and depth can be derived from disparity given the cameras' focal length and baseline.

Methods range from local to global. Local methods, such as block matching, compare small windows to find best matches but can produce noisy results in textureless regions. Global and semi-global approaches enforce smoothness constraints across the image, yielding more accurate disparity maps but with higher computational cost. Modern systems may fuse stereo data with other sensors for robustness.

Applications include autonomous vehicles, robotics, 3D reconstruction, medical imaging, and virtual or augmented reality. Stereovision enables dense depth maps and 3D models without active sensing, such as structured light or LiDAR, though it depends on texture, lighting, and baseline geometry.

Limitations include occlusions, lack of texture, repetitive patterns, reflective or transparent surfaces, and sensitivity to calibration errors. Real-time performance requires optimized algorithms and hardware. Despite challenges, stereovision remains a foundational approach for acquiring depth information.