I-Corps: Low False Negative 3-D Facial Recognition

I-Corps项目更广泛的影响/商业潜力是将面部识别技术扩展到发射红外波段，并在存在较大面部姿势变化的情况下减少假阴性。基于可见二维图像的机器学习技术在商业面部识别领域占据主导地位，而异构面部识别技术，包括三维到二维匹配和红外到可见匹配，还没有看到类似的广泛的商业部署。这项专利技术来自美国国家科学基金会奖，为商业部署带来了关键的算法组件，增强了异构面部识别技术，同时不依赖于机器学习及其伴随的大型训练数据库要求。这反过来又打开了针对给定监视名单的改进自动监视的潜力，这些监视名单可部署在许多公共和私人空间中，并且在比目前使用2d商业匹配算法更广泛的监视条件下。这个I-Corps项目是在美国国家科学基金会（NSF）奖励下开发的技术成果，通过减少面部隐式表面生成和多光谱（可见和红外）源的生物特征捕获的计算成本，可以彻底改进3-D面部识别。该技术利用三维面部隐式曲面对存在较大姿态变化和遮挡的面部进行鲁棒识别。在自动监控的典型条件下，现有的二维商业面部识别算法的改进是显而易见的，在这种情况下，使用传统的做法，面部姿势变化大，光线不足会导致假阴性。然而，作为本研究基础的3d面部识别方法和设备，在面部姿势变化较大的情况下表现同样良好，并且能够将多波段红外输入与可见图像输入无缝集成，即使在低光和夜间条件下，也能大大降低假阴性面部识别。该奖项反映了美国国家科学基金会的法定使命，并通过使用基金会的知识价值和更广泛的影响审查标准进行评估，被认为值得支持。

The broader impact/commercial potential of this I-Corps project is the expansion of facial recognition technologies into the emissive infrared bands and the reduction of false negatives in the presence of large facial pose variation. Visible 2-D image-based machine learning techniques dominate the commercial facial recognition landscape while heterogeneous facial recognition techniques including 3-D to 2-D matching and infrared to visible matching have not seen similar widespread commercial deployment. The patented technology originates from an NSF Award and brings a crucial algorithmic component for commercial deployment that enhances heterogeneous facial recognition techniques while not relying on machine learning nor its concomitant large training database requirements. This in turn opens up the potential for improved automated surveillance against a given watch-list deployable in a host of public and private spaces and under a wider range of surveillance conditions than currently possible using 2-D commercial matching algorithms.This I-Corps project is a result of technology developed under an NSF Award which enables a radical improvement in 3-D facial recognition by reducing the computational costs for facial implicit surface generation and biometric capture originating from multispectral (visible and infrared) sources. The technology results in robust recognition in the presence of large facial pose variation and occlusion using 3-D facial implicit surfaces. The improvement over existing 2-D commercial facial recognition algorithms is evident under conditions typical of automated surveillance where large facial pose variation and poor lighting result in false negatives using conventional practice. The method and apparatus for 3-D facial recognition underlying this research, however, performs equally well under large facial pose variation and is able to seamlessly integrate multiple band infrared input along with visible image input for vastly lower false negative facial recognition even in the presence of low-light and nighttime conditions.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.