Extraction of Light Stripe Centerline Based on Self-adaptive Thresholding and Contour Polygonal Representation

Abstract

Extracting light stripe centerline is the key step in the line-structure light scanning visual measuring system. It directly determines the quality of three-dimensional point clouds obtained from images. Due to the reflectivity and/or color of object surface, illumination condition change and other factors, gray value and curvature of light stripe in image will vary greatly that makes it very difficulty to completely and precisely extract sub-pixel centerline. This paper presents a novel method for light stripe centerline extraction efficiently. It combines the integral image thresholding method, polygon representation of light stripe contour and adaptive center of mass method together. It firstly locates light stripe region and produces binary image no matter how change gray values of light stripe against background. Then the contour of light stripe is extracted and approximately represented by polygon. Based on the local orthogonal relationship between directions of light stripe cross-section and corresponding polygon segment, the direction of light stripe cross-section is calculated quickly. Along this direction, sub-pixel centerline coordinates are calculated using adaptive center of mass method. 3D scanning experiments with human model dressed colorful swimsuit on a self-designed line laser 3D scanning system are implemented. Some comparisons such as light stripe segmentation using 3 thresholding methods, the time used and the smoothness are given and the results show that the proposed method can acquire satisfying data. The mean time used for one image is not beyond 5 ms and the completeness and smoothness of point clouds acquired by presented methods are better than those of other two methods. This demonstrates the effectiveness and practicability of the proposed method.