CRI: A Core Experimental Facility for Computer Vision and Artificial Intelligence

摘要提议：CNS 0454056PI：卡洛·托马西研究所：杜克大学项目：NSF 04-588简明计算研究基础设施CRI：计算机视觉和人工智能的核心实验设施杜克大学将建造一个测量人体三维形状和运动的设施，用于计算机视觉和人工智能算法的培训和性能评估。形状将使用高质量的激光测距仪进行测量。同一场景或物体的结构光、立体和单目图像也将从不同质量的传感器系统中收集，其相对于激光测距仪的位置和方向通过校准准确地知道。同样，一个或几个人在太空中的运动将通过精确的运动捕捉系统进行记录，以及几套彩色和黑白相机和结构光系统同时记录。计划进行立体视觉、机器人定位、可视化、手势分析和皮肤学的研究项目。激光测距仪和运动捕捉系统提供准确的训练和地面真实数据，其他传感器产生计算机视觉算法或人工智能推理系统用于确定或推理场景或事件的输入类型。由于对这类数据的广泛需求，该提案还设想开设一门关于传感和几何测量的基本原理的课程，其材料也将予以公布，并为在该设施和其他地方收集的测量数据建立一个网上储存库。

AbstractProposal: CNS 0454056PI: Carlo TomasiInstitution: Duke UniversityProgram: NSF 04-588 CISE Computing Research InfrastructureCRI: A Core Experimental Facility for Computer Vision and Artificial Intelligence Duke University will construct a facility for the measurement of three-dimensional shapes and the motion of human bodies, to be used for both training and performance evaluation of computer vision and artificial intelligence algorithms.Shapes will be measured with a high-quality laser range finder. Structured light, stereoscopic and monocular images of the same scene or object also will be collected from sensor systems of varying quality, and whose positions and orientations relative to the laser range finder are precisely known through calibration. Similarly, the motions in space of one or a few people will be recorded with an accurate motion capture system, as well as with several sets of color and black-and-white cameras and structured-light systems recording at the same time. Research project on stereo vision, robot localization, visualization, gesture analysis and dermatology are planned.The laser range finder and motion-capture system provide accurate training and ground-truth data, and the other sensors yield the type of input that a computer vision algorithm or an artificial intelligence inference system would use to determine or reason about a scene or an event. Because of the broad need for this data, the proposal also envisions the creation of a course on the principles underlying sensing and geometric measurements whose materials will also be made public, and of a web repository for the measurement data collected both at the facility and elsewhere.