Discreet Facial Recognition on a Standalone Mobile Platform with Augmented Reality Feedback

• Main Author: Stockton, Patrick Michael
• Format: Dissertation
• Language: English
• Published: ProQuest Dissertations & Theses 01-01-2020
• Subjects: Artificial intelligence; Computer Engineering
• ISBN: 9798641755670

The concept of machine learning and computer vision has been established for the last few decades, however their viability in everyday usage has not been practical until recently. With the emergence of powerful enough devices and software tools, the viability of using machine learning and computer vision has been realized in a near infinite number of applications. One such popular application is real-time facial recognition. As facial recognition is one such application, the context in which it is needed and used leads to the ability to offer such a solution to a wide number of users. These solutions can include recognition of patients in a hospital, students in a classroom, allowing individuals suffering from dementia to recognize certain faces, residents in nursing homes to recognize each other if their names are forgotten, and many others. The ability to keep such a discreet, offline, and pocket-size platform is critical for the comfortable and convenient use in everyday situations. To achieve the requirement of a small microcomputer system that can perform face recognition tasks without the need of a larger computer, the use of a modern computing solution is essential. The use of augmented reality smart glasses provides the user with the real-time information from the mobile computing system discreetly on its heads-up display. This platform is discussed in this research along with its implementation. This research presents the design of such a mobile computing system that can perform real-time facial recognition while providing discreet feedback to the user through an augmented reality display. The NVIDIA Jetson Nano serves as the computing platform that performs face classification on an input video stream using a custom face image dataset. This custom dataset was trained on two face recognition models and machine learning libraries for performance comparison. When a face is detected in the video stream the system will send the information to the augmented reality smart glasses to discreetly display the pertinent information to the user.