978-1-4244-5023-7/09/$25.00 ©2009 IEEE September 14-16, 2009 METU Northern Cyprus Campus 1 Single Stripe Projection based Range Scanning of Shiny Objects Under Ambient Light Rıfat Benveniste and Cem ¨ Unsalan Computer Vision Research Laboratory Department of Electrical and Electronics Engineering Yeditepe University ˙ Istanbul, 34755 TURKEY e-mail:unsalan@yeditepe.edu.tr Abstract—Range scanners are used in various applications in industry. Therefore, various range scanners, based on different working principles are developed. Among these, scanners using stripe based triangulation are the most promising ones. Unfor- tunately, these scanners have problems in obtaining the range data of shiny objects (having highlights) under ambient light. In this study, we develop a stripe projection based range scanner to solve this problem. Our main contribution is on detecting the stripe using a color invariant in a robust manner. We developed a projector based scanner using this color invariant. Our scanner has also the advantage of flexibly changing stripe color for different colored objects. We test the performance of our novel stripe detection method under various controlled experiments. We also test the overall performance of our scanner in extracting the range data of several objects having highlights. Extensive testings indicate the success of our range scanner, especially for scanning objects having highlights under ambient light. I. I NTRODUCTION Range scanners are used in various areas such as reverse engineering, quality inspection, robotics, computer graphics, cultural heritage (archeological finding), and medical imaging. Some applications on these areas can be counted as: modeling and recovering broken or decayed machine parts, quality inspection, 3D animated character generation, orthopedical and reconstruction surgery model preparations and digitally archiving of objects from cultural heritage (like virtual muse- ums). The first step for all these applications is acquiring the range data of the object in a robust manner. Therefore, several range scanner systems based on different operating principles are introduced. A widely used scanner type with high accuracy is based on stripe triangulation [3]. In this system, the stripe is projected onto the target object from a known direction. The projected stripe takes the shape of the object. A camera which is spatially displaced wrt. the stripe source captures the scene. The real world 3D object coordinates are calculated by computing the intersection of the camera’s line of sight with the stripe plane. The important step here is reliably locating the stripe projected onto the object surface in the acquired image [15]. The simplest approach here is to scan each row of the image to detect peaks in intensity by locating the position of the first or largest response. Therefore, most scanners use line lasers (having high illumination intensity) as a stripe projecting source. It is easy to detect peaks on the acquired image as the location of the stripe coming from a powerful light source like laser. However, scanned object’s surface properties and the ambient lighting conditions may easily affect peak detection. Especially ambient light creates extra highlights on the objects with shiny surfaces. These highlights lead to false range measurements besides the actual range data. Some commercial range scanners warn their customers about this shortcoming. Several researchers also mention the problem of scanning shiny surfaces by a laser stripe based range scanner [1], [2], [4], [9], [10], [16]. Most of these researchers also mentioned that, there is a strong need of robust data acquisition for challenging surfaces (like the ones having high reflectance). There may be simple solutions like painting or covering the surface of the shiny object with a powder. However, as Park and Kak [14] mentioned, this may not be feasible for most applications. A typical example may be given on archeological findings. The paint or the powder may contaminate the object surface. One may also think that scanning under dark can solve problems originating from ambient light. Especially for outdoor scanning (like scanning statues), the illumination level may not be controlled. As in robotics applications, the illumi- nation level may also change during the scanning operation. To handle all these problems, a robust range scanning system is needed that can work under different ambient illumination levels. In the literature, there are various attempts to solve the prob- lem of scanning shiny objects under ambient light. Umasuthan and Wallace [18] tried to solve the problem from the obtained range data. They used a least squares estimator to remove outliers in the range data. Elgazzar et al. [5] developed a specific laser stripe based range sensor for indoor environment scanning. They, modified the lens of the camera with a mask in front of it. They counted the insensitivity to ambient light as one of the advantages of this setup. Levoy et al. [13] tried different lighting conditions to decrease the effect of laser stripe scattering. Singhal et al. [6] introduced a technique to eliminate spurious range values using two (or more) cameras and several consistency tests. Forest et al. [7] proposed an