3D shape measurement of human shape from multiple viewpoint images using shape database

本研究的目的是利用由数百个人体部位的三维形状组成的数据库，建立一种从多视点图像中测量精度接近1 mm的人体部位的三维形状的方法。在本研究中，我们首先探索了一种从固定和校准的多个摄像头测量食物形状的方法，然后开发了一个能够方便地重建足部形状的试点系统。通过对该方法的扩展，我们还建立了一种从获取面部/头部区域的多个视点图像的图像序列中恢复人脸/头部区域形状的方法。除了这一成果之外，我们还提出了一种从正视图像中重建人脸形状的方法。最后，我们将这些方法扩展到一种从未标定和不同步的多个摄像机拍摄的多视点图像中恢复人脸形状的方法，其中我们假设目标人脸被广泛用于安全目的的监视摄像机捕获的情况。作为结果，我们验证了所提出的方法的有效性：1)从多个标定的摄像机重建食物形状的方法，2)由手持摄像机自由移动的多视点图像重建面部/头部形状的方法，3)利用放置在相机的相同位置的单点照明从面部正面捕获的单个图像重建面部/头部区域形状的方法，以及4)利用放置在相机的相同位置的单点照明从面部正面捕获的单个图像重构面部/头部区域形状的方法，以及5)从用未校准和不同步的多个相机捕获的多个视点图像重构面部/头部区域形状的方法。

The purpose of this research project is establish a method for measuring 3D shapes of human parts with approximate 1mm accuracy from multiple viewpoint images by using database that consists of several hundreds 3D shapes of human parts.In this research, we first pursue a method for measuring food shape from fixed and calibrated multiple cameras, and then have developed a pilot system that can easily reconstruct the shapes of foot shapes. By extending this method, we also established a method for recovering the shape of human facial/head area from an image sequence that captures multiple viewpoint images of the facial/head area with a handy camera that is moved freely by hand. In addition to this achievement, we also proposed a method for reconstructing facial shape from a frontal view image. Finally we extend those methods to a method for recovering facial shape from multiple viewpoint images taken by uncalibrated and unsynchronized multiple cameras, where we assume a situation that a object human's face is captured by surveillance cameras that are widely used for security purpose.As results, we demonstrated the effectiveness of the proposed methods : 1) the method for reconstructing food shape from multiple calibrated cameras, 2) the method for reconstructing facial/head area shape from multiple viewpoint images that are captured with a handy camera moving freely with a hand, 3) the method for reconstructing facial/head area shape from a single image captured from the front of the face with a single point illumination placed at the same position of the camera, and 4) the method for reconstructing facial/head area shape from a single image captured from the front of the face with a single point illumination placed at the same position of the camera, and 5) the method for reconstructing facial/head area shape from multiple viewpoint images captured with uncalibrated and unsynchronized multiple cameras.