Abstract: In order to detect the edge misalignment of rail weld in real time, an automatic positioning method for the rail weld based on computer vision is proposed. First, the method of median filtering is used to remove the noise in a weld image. Second, the weld image is enhanced by using the contrast limited adaptive histogram equalization (CLAHE) algorithm and the histogram equalization method. Third, the image is segmented by the double threshold OTSU method to highlight the rail head and the weld region, and the contours of the rail head and the weld region are extracted by using the method of connected-domain. Finally, the weld fitting line is obtained by multiple least squares fitting, and the intersection of the fitted straight line and the top edge line of the rail head are calculated as the positioning point, accordingly, the rail weld is automatic positioned. Through a test of the weld region of 60 kg/m rail, the method performs with high precision and stability, and can be used for measuring the rail weld automatically in real-time on line.

Key words: computer vision, image enhancement, weld edge detection, automatic positioning