Corisco: Robust edgel-based orientation estimation for generic camera models

• We propose an orientation estimation method based on edgels and M-estimation.
• Edgels are extracted with a grid mask that can compromise between speed and accuracy.
• Any camera model can be used, and errors are calculated on the original image space.
• The estimation starts with a random search and ends with a continuous optimization.
• The method uses quaternions, and all derivative calculations use closed formulas.

The estimation of camera orientation from image lines using the anthropic environment restriction is a well-known problem, but traditional methods to solve it depend on line extraction, a relatively complex procedure that is also incompatible with distorted images. We propose Corisco, a monocular orientation estimation method based on edgels instead of lines. Edgels are points sampled from image edges with their tangential directions, extracted in Corisco using a grid mask. The estimation aligns the measured edgel directions with the predicted directions calculated from the orientation, using a known camera model. Corisco uses the M-estimation technique to define an objective function that is optimized by two algorithms in sequence: RANSAC, which gives robustness and flexibility to Corisco, and FilterSQP, which performs a continuous optimization to refine the initial estimate, using closed formulas for the function derivatives. Corisco is the first edgel-based method able to analyze images with any camera model, and it also allows for a compromise between speed and accuracy, so that its performance can be tuned according to the application requirements. Our experiments demonstrate the effectiveness of Corisco with various camera models, and its performance surpasses similar edgel-based methods. The accuracy displayed a mean error below 2° for execution times above 8 s in a conventional computer, and above 3° for less than 2 s.