Kearfott K J, Juang R J, Marzke M W
G. W. Woodruff School of Mechanical Engineering, Georgia Institute of Technology, Atlanta 30332-0405.
Med Biol Eng Comput. 1993 Mar;31(2):149-56. doi: 10.1007/BF02446673.
The surface structure of the growing portion of bones, called the metaphysis, contains clues about the locomotor characteristics of various species. Present methods of capturing this anthropologically interesting surface are time-consuming and subject to human error. The research implements a digital stereo imaging technique for bone metaphyses and joints in skeletal collections. The corresponding points in two images collected from different angles are determined using an area-based correlation matching method. The depths of matched points are computed from the difference in location of the points in the two images. The paper presents a practical implementation of computer vision for anthropology using an 80286-based personal computer, a camera and a video digitiser. The stereo matching algorithm, a practical implementation of classical stereo imaging methods, takes less than 1 min and produces reasonable representations of mammal bones. The accuracy of the depth measurements ranged from 0.7 to 12 per cent for 45-150 cm object-camera distances. False matches occurred in approximately 6 per cent of the total matched points.
骨骼生长部位的表面结构，即干骺端，包含了有关各种物种运动特征的线索。目前获取这种具有人类学研究价值的表面的方法既耗时又容易出现人为误差。该研究为骨骼标本中的干骺端和关节实施了一种数字立体成像技术。使用基于区域的相关匹配方法来确定从不同角度采集的两幅图像中的对应点。匹配点的深度通过两幅图像中这些点位置的差异来计算。本文介绍了一种利用基于80286的个人计算机、一台相机和一台视频数字化仪实现的用于人类学的计算机视觉实用方法。立体匹配算法是经典立体成像方法的一种实际应用，耗时不到1分钟，能生成哺乳动物骨骼的合理图像。对于45 - 150厘米的物体与相机距离，深度测量的精度范围为0.7%至12%。在总匹配点中，大约6%出现了错误匹配。
