基于曲面深度学习的高速三维人脸测量技术研究

3D concept becomes popular and has started penetrating into our daily life. It is highly need to develop high-speed 3D face measurement technology that can be used for many places (e.g., railway stations and airports) for biometrics, surveillance etc. However, none of the-state-of-art 3D facial geometry capture technology can do that because hardware limitations, and partially the mathematically tools used in the 3D shape measurement field is not advanced enough.. To conquer the challenges of high-accuracy and high-speed 3D shape measurement techniques, this research endeavors to bring advanced mathematical tools including inner surface geometric properties and deep learning method to create a novel 3D facial measurement framework.. We propose to develop an 3D shape measurement platform based on a known geometric shape object (e.g., a ping-pong ball); then use the advanced Riemann geometry tools to analyze the inner geometric properties of the face itself (e.g., the distance between key points, and the rigid transformation of some region of face texture); and finally employ deep learning technology and the Wasserstein metric using optical transport map to create an entire new high speed 3D face measurement platform.. The PIs have broad mathematics knowledge and rich experiences on digital fringe project technology for many years. If successful, this project not only can solve the difficult problem on 3D face measurement, but also create a new research direction between interdisciplinary subjects, and make substantial contributions to the optical metrology field as well as to our society.

随着三维的概念逐渐深入人心，人们对非合作的高速三维人脸测量技术有了非常迫切的需求，如车站机场等，然而该技术却一直是研究难点，除了硬件上的限制外，部分原因是数学工具不够先进导致算法不够成熟。.为了克服这一困难，本项目力求从数学角度进行分析，以人脸曲面自身的内蕴几何特征为基础，通过深度学习等手段，建立全新的高速三维人脸测量理论框架。.项目首先提出基于已知几何的参考物的测量算法平台；以此为基础，根据人脸纹理关键特征点在运动中保持距离不变以及部分特征区域运动几乎不变形的特性，又提出了两套新方案；更进一步，基于三维特征曲面的深度学习，以及利用最优传输映射建立Wasserstein度量，项目建立了基于内蕴几何的一整套高速人脸测量技术的新方法。.申请者拥有丰富的数学背景知识和在数字条纹投影技术方面的多年经验，如成功，可以一举解决这个多年难题，开辟出一条全新的交叉学科研究方向，为我国科研事业做出贡献。

高速高精的三维测量技术一直是光学测量领域的研究热点，其中的基于数字条纹投影技术，因为其稳定性和精确性，在高速三维测量领域广受关注。然而通常的三步相移技术只能得到相对相位，使得高速高精的三维测量技术发展遇到了瓶颈。.为了解决上述问题，本项目分析了高速高精三维测量过程中的三大关键难点，提出并完成三个核心算法，分别是：基于连续性的多个离散物体高速高精测量算法，该算法根据连续性调整最优的最小参考平面完成离散物体的高度高精测量；基于双参考平面的大景深测量算法，该算法利用对两个参考平面之间的重叠区域进行分析以及分类，从而达到正确的绝对相位展开来完成测量；以及基于区域生长的遮挡物体测量技术，该算法无需其他额外的信息，仅仅是利用不同的相位区域进行分析重组，从而达到相位解包裹的目的。.项目开发过程中，项目组也根据购买的设备，进行了系统的外包装设计以及定制，并成功地搭建出一套可以便捷使用的工业测量系统。.这些核心算法以及可操作的测量系统平台，这些都将为相关的各行业服务。