本文为比利时列日大学（作者：Antoine Dubois），共130页。

创新一直是人类文明发展的动力，通过引入新的思想来解决这样一个复杂系统会遇到无数的问题。创新刺激了经济增长，给我们今天的生活带来了方便。在我看来，在比利时，特别是在南部，我们常常停留在功劳上，而忘记了创新。这一观察结果促使我选择在RAGI项目的背景下进行本论文的研究。这个创新项目依靠学术研究来设计商业产品，开发一个“智能识别、欢迎与引导系统”。该项目的创新部分来自于使用一种称为自动人脸识别的机器学习技术，能够识人类并在建筑物中引导他们。本文的目的是研究这一概念，以帮助从事RAGI工作的团队做出适当的决策。为了实现这一目标，本文从人脸检测和人脸识别两个方面对算法进行了研究分析。算法研究通过分析最近的基准，其中两篇文献（即WIDER FACE [111]和MegaFace [47]）也用于评估这些算法。同时，本文也指出了进行这种研究和测试过程的困难。这些基准测试的结果可以确定哪些算法性能更好，也就是说SSH[65]用于检测，Dlib-R[20]和ArcFace[21]用于识别。为了进行检测，研究了姿态、大小或模糊等人脸属性对检测结果的影响。最后，为了得到更多与RAGI项目相关的结果，我们设计了一个特定的数据集，在这个数据集上测试了相同的算法。它由494个帧和3561个来自13个不同身份的带注释的面部组成，允许我们在确认的公开数据集上获得结果，同时研究其他参数。所有这些测试都是在考虑算法效率的情况下进行的，计算时间测试表明，性能最好的技术往往在速度上比较慢。

Innovation or the introduction of the new.Innovation has always been the motor of civilization by introducing new ideasto solve the countless problems that such a complex system encounters. It hasspurred economic growth and brought the comfort we live in today. In myopinion, in Belgium, particularly in the south, we have often rested on ourlaurels and forgot to innovate. This observation fostered my choice toundertake this thesis in the context of the RAGI project. This innovativeproject, relying on academic research to produce a commercial product, isdeveloping an “intelligent system for recognition, welcoming, and guidance”.Innovation in this project comes partly from the use of a machine learningtechnique called automatic face recognition to be able to identify people andguide them in a building. The goal of this thesis is to study this concept tohelp the team working on RAGI take appropriate decisions. To achieve this goal,I search for and analyze state-of-the-algorithms in both face detection andface recognition. The research for algorithms goes through the analysis ofrecent benchmarks, two of which (i.e. WIDER FACE [111] and MegaFace [47]) arealso used for evaluating those algorithms. Simultaneously, this work points outthe difficulties to perform such a research and testing process. The results onthese benchmarks allows to determine which algorithms perform better, that isto say SSH [65] for detection and both Dlib-R [20] and ArcFace [21] forrecognition. For detection, the influence of facial attributes such as pose,size or blur is explored. Finally, to have more relatable results with regardsto the RAGI project, we designed a specific dataset on which the samealgorithms are tested. Composed of 494 frames with 3561 annotated faces from 13different identities, it allowed us to study other parameters while confirmingthe results obtained on the publicly available datasets. All those tests areperformed with algorithm efficiency in mind and computation time measurementsshow that the best techniques tend to work slower but that they can achievepractical execution times.

1. 引言

2. 项目背景

3. 基准分析

4. 算法选择

5. 算法描述

6. 算法评估

7. RAGI项目上的应用

8. 结论

附录 数据集样本

更多精彩文章请关注公众号：