题目/Title：Capturing Texture for robotic Fingers using 3-D reconstruction

作者/Author：闫岩,张春,朱曦阳
                        Yan Yan,Chun Zhang,Xiyang Zhu

会议/Conference：ICCSIP 2016

地点/Location：Beijing, China

年份/Issue Date：2016.19-23 Nov

页码/pages：

摘要/Abstract：
Sense of touch is vital for robotic hands when performing complicated dexterous movements. Texture perception is an essential part of the sense. In this paper, we describe a system which obtains surface texture by means of 3-D reconstruction. The system contains a sensor, four evenly spaced LEDs and a camera. The sensor is a slab of silicone which is covered with an aluminum skin made by semiconductor sputtering on one side. When pressed, it will distort to record the shape of the surface texture. Every time we capture an image with only one LED illuminating in one direction and this provides four images. With the surface structure becoming complicated, shadows are unavoidable. Our algorithm is able to deal with shadows: for areas which are not in shadow, as the metal skin offers the same bidirectional reflectance distribution function (BRDF) for all objects, we use photometric stereo; for shadow regions, a dimension reduction search algorithm is used to get an accurate surface normal vector. Finally, we obtain reliable depth information from gradients and reconstruct surface texture, which helps give robotic fingers the sense of touch.