A Method to Detect Corner Points and Feature Lines Based On Edges in Images in Visual SLAM

Points and feature lines correctly even in different brightness and noise conditions. With the purpose of making a contribution to the improvement of the performance of ORB (Oriented FAST and Rotated BRIEF) corner detection which is widely used in SLAM system, this paper proposes a method to determine corner points and feature lines based on gradient direction changes calculated at the edge points detected in the image. First, the curve of edge points is represented by a multivalued function to be used as information about feature lines. Then, a method to detect corners on edges by examining the change in the gradient direction at the edge points, not by calculating curvatures of the edge point curve, is proposed. The part of the edge curve represented by a multivalued function near corners has been emphasized to be used as information about feature lines. Finally, the proposed method was tested with EuRoC dataset images in terms of stereo camera visual SLAM and the test results show that the proposed method detected corner points better than edge-based corner detectors based on point-to-centroid distance(PCD), Simple Triangular Theory and Curvature Scale Space(CCS), and ORB.