Development of a Pressure-Based Gait Biometric for User Access Control

在当今的数字时代，人员身份识别变得越来越重要，物理和网络威胁变得司空见惯。深度学习的快速进步导致基于视频的面部识别系统在机场、医院和购物中心等公共场所的开发和采用。尽管如此，仍然存在巨大的挑战，包括照明条件、视角以及由于服装、个人防护设备或拥挤的空间而导致的视觉遮挡。最近，这些基于视频的系统也受到了审查，因为诉讼指控侵犯隐私，并且算法中固有的偏见（由于不公平的训练过程）已被识别。 这使得执法部门、安全机构和行业寻求新的用户访问控制解决方案。作为面部识别的替代方案，步态识别已经出现，它通过人们走路的方式来识别他们。研究导致基于视频的步态识别的快速改进，再次利用深度学习来提取步态的骨骼模型。然而，对基于愿景的模式的依赖意味着许多相同的挑战和担忧也适用。与基于视频的步态识别相比，基于传感器的方法（例如通过压敏地板实现的方法）消除了视频捕获的需要，减轻了对图像重复使用的担忧，并避免了照明、可见度、外观变化和遮挡的不利影响。在这项工作中，我们将与 Stepscan Technologies 和 Cyber​​NB 合作，在一座新的高安全性门禁建筑中安装一种新型压敏地板系统，通向一组刷卡激活旋转栅门。通过这个装置，我们将开发一种新颖的基于压力的步态识别系统，用于用户身份验证，并进行纵向的真实世界评估。结合受控的实验室实验来探索和解决影响该系统性能的混杂因素，该项目将生成用于基于压力的步态分析的首个数据集、有关步态生物识别及其混杂因素的新知识、人类运动机器学习的进步以及为学员和行业合作伙伴提供转化学习体验。

Person identification is increasingly important in today's digital age, with physical and cyber threats becoming commonplace. Rapid improvements in deep learning have led to the development and adoption of video-based facial recognition systems in public areas such as airports, hospitals, and shopping malls. Still, substantial challenges remain, including lighting conditions, perspective, and visual occlusion due to clothing, personal protective equipment, or crowded spaces. Recently, these video-based systems have also come under scrutiny as lawsuits have alleged invasion of privacy, and as inherent biases in the algorithms (due to inequitable training processes) have been identified. This has left law enforcement, security agencies, and industry seeking new user access control solutions. As an alternative to facial recognition, gait recognition, which involves identifying people by the way they walk, has emerged. Research has led to rapid improvements in video-based gait recognition, again leveraging deep learning to extract skeletal models of gait. The reliance on vision-based modalities, however, means that many of the same challenges and concerns apply. Compared to video-based gait recognition, sensor-based methods (such as those enabled by pressure-sensitive flooring) eliminate the need for video capture, mitigating concerns over image reuse and avoiding the adverse effects of illumination, visibility, appearance changes, and occlusion. In this work, in partnership with Stepscan Technologies and CyberNB, we will install a novel pressure-sensitive flooring system leading up to a set of swipe card activated turnstiles at a new high-security access-controlled building. With this installation, we will develop a novel pressure-based gait recognition system for user authentication, and conduct a longitudinal, real-world evaluation. Combined with controlled in-lab experiments to explore and address confounding factors to the performance of this system, this project will generate a first-of-its-kind dataset for pressure-based gait analysis, new knowledge about gait biometrics and its confounding factors, advancements in machine learning of human movement, and a translational learning experience for trainees and industry partners.