Study On Line Detection Algorithms Based On Principal Direction And Curvature

Automatic line detection is widely used in robot navigation, ship celestial navigation, camera calibration, industrial inspection, character recognition and other fields.Hough transform is a classic line detection algorithm. It uses the edge operator to detect the image edge, then uses the Hough transform to vote, and lastly obtains the line parameters by detecting peaks in the parameter space. Commonly used edge detection operators include Canny operator, Sobel operator, Laplace operator, etc. Hough transform commonly uses Canny operator as the edge detection operator.In this thesis, a line detection algorithm based on principal direction is proposed. First, by treating gray images as two-dimensional discrete surfaces, principal curvatures and principal directions of the given pixel can be approximately calculated. Second, since the minimum principal direction of the pixel on the line indicates the direction of the line, the consistency of principal direction of lines can be used to detect lines. Finally, potential line pixel selection and pixel shifting are employed for acceleration and accuracy. Both theoretical analysis and simulation data indicate that the line detection algorithm based on principal direction has advantage over Hough transformation, LSD, and EDlines in time consumption. These four methods cost 17.35ms,182.10ms,177.82ms, and 17.71ms, respectively, to detect a same image under the same environment.This thesis also presents a line detection algorithm based on Hough transform. Edge pixel is extracted based on the principal curvature. The voting count is reduced by probabilistic sampling of the edge pixels. One-to-one voting is enforced on the information of the main direction. Experimental results show that the presented method has better positioning accuracy and computational efficiency than other HT algorithms.