Fast Range-lmage Generator by using Gray coded pattern projection

本计画主要探讨深度图的获取方法，以及深度图在三维物体识别上的应用。我们已经开发了一种新型的距离成像系统，使用图案投影。基于三角法的空间编码测距系统由图形投影仪和CCD摄像机实现。设计了采用格雷码光和半平面光二值图像积累的实验系统。我们有实验结果。前一种方法不需要一个狭缝光依次扫描2n个位置，只需几个（n）点的光投射就能稳定、快速地得到精确的距离像。后一种方法实现了用非常简单的设备构造的测距仪。这些系统可以应用于工业上的形状检测、形状测量和机器人视觉等领域。为该系统制作的实验图形投影仪采用机械光罩结构。非机械图案投影器，例如LCD（液晶器件）快门，希望用于工业用途。对点阵式液晶显示器进行了初步的实验。实验结果表明，距离图像的投影对比度较好，电极间隙的影响可以忽略不计，距离图像可以方便地进行场景分析，包括三维物体的分割、描述和理解。为了识别3D对象的形状，已经研究了以下方法。应用二次曲面逼近将曲面分为椭球面、抛物面、圆锥面等几类。环形算子和曲率算子可以检测深度图像中的特征点。

This project deals with methods to get range picture and applications and utilizations of range picture for Three-dimensional(3D) object recognition. We have developed new type of range-imaging systems using pattern projection. Pattern projector and CCD(Charge Coupled Device) video camera realize the range measurement system utilizing space encoding based on triangulation method. Both the experimental system using Gray coded pattern ligh, and one using binary image accumulation with half-plane light, are produced. We have experimental results. The former approach gets accurate range picture stably and quickly with a few (n) of light projection, instead of scanning a slit light at 2n positions in turn. The latter approach realizes range finder constructed with very simple equipments. It is confirmed that these systems may be suited for the industrial use, for example, shape inspection, shape measurement and also robot vision.Gray coded pattern projection method requires the switching of light-masks. Experimental pattern projector produced for the system have mechanical light-mask structure. Non-mechanical pattern projector, such as LCD(Liquid Crystal Device) shutter, are wished for industrial use. Dot matrix LCD display is experimented elementally. It is confirmed that the contrast of projection is considerable good and the influence of gap of electrode can be ignored.Range picture are very convenient for scene analysis including segmentation, description, and understanding of 3D objects. In order to recognize the shape of 3D objects, the following methodologies have been studied. Quadric surface approximation is applied to classify the spape of surface into some categories, for instance, ellipsoid, paraboloid and cone. Ring operator and curvature operator can detect feature points in range picture.