Multiple contour sequences' segmentation and entity recognition methods in vision measurement

Abstract

ABSTRACT In this paper, an approach is proposed for segmentation of multiple contour sequences and recognition of entities for vision measurement of small precision pa rts. The approach includes several step s as follows. All contour sequences of the part are detected at the first place. S econdly, a circle identification method is us ed to find circular contours in contour sequences. The identified circular contours are further fitted as individual circles. Then, curvature method is selected to detect dominant points in the rest contou rs and height projection method is adopted to classify them as line or arc entities. In the end, the least-squares method is used to merge and add dominant points. Experimental results show lines, arcs and circles can be recognized satisfactorily by using the approach presented. Keywords: Vision measurement, contour segmentation, entity recognition, dominant point detection, least-squares method 1. INTRODUCTION Segmentation of contour sequences and recognition of entities on a two-dimensional part are essential tasks in vision measurement, pattern recognition, shape analysis and reverse engineering. Segmented contours and identified entities can result a meaningful and compact description of a part. Such description is useful for higher level vision processing. In vision measurement, contour segmentation and recogniti on can impact on the reliability and accuracy of dimensions measured. In the past two decades, many techniques have been proposed for this purpose. Polygonal approximation is one approach where the line segments are extracted from curves based on the corner or dominant point detection