Optimal line feature generation from low-level line segments under RANSAC framework

The low-level line segment features have low accuracy as they are more easily affected by the noise and differeent line segment detectors. Furthermore, the line segment is not a good feature for matching across multiple views when we need to finish the 3D reconstruction. However, the line feature is more robust for the noise. In this paper, a kind of line feature, ideal line, is defined and optimally estimated by clustering the line segments under RANSAC framework. The physical experiments are carried out to verify the proposed estimation method.