A Minimal Solution for Image-Based Sphere Estimation

Abstract We propose a novel minimal solver for sphere fitting via its 2D central projection, i.e. , special ellipse. The input of the presented algorithm consists contour points detected in camera image. General ellipse problems require five points. However, taking advantage isotropic spherical target, three are enough to define tangent cone parameters sphere. This yields sought parameters. Similarly, center can be estimated from if radius is known. These proposed geometric methods rapid, numerically stable, and easy implement. Experimental results—on synthetic, photorealistic, real images—showcase superiority solutions state-of-the-art methods. A real-world LiDAR-camera calibration application justifies utility sphere-based approach resulting an error below few centimeters.