当研究所の研究員である鈴木 芽衣、野本 一輝が執筆した研究論文が、情報処理学会論文誌に採録されました。
PinchKey: A Highly Accurate and Usable Behavioral Biometric Authentication Method for VR Users
著者：Mei Suzuki (Waseda University/DTCY), Ryo Iijima (Waseda University/AIST), Kazuki Nomoto (Waseda University/DTCY), Tetsushi Ohki (Shizuoka University/RIKEN), Tatsuya Mori (Waseda University/RIKEN/NICT)
概要：This study introduces PinchKey, a highly accurate and user-friendly behavioral biometric authentication method for VR users. Authentication is performed using the natural and intuitive ‘pinch’ gesture used when operating VR devices. Since this two-finger action is a natural and intuitive gesture used when operating VR devices, we expect that the use of the pinch gesture will provide enhanced usability compared to traditional VR device authentication methods such as pattern lock or PIN entry. The system analyzes the three-dimensional trajectory of the user's finger movements during the pinch gesture and applies a supervised machine learning model to perform authentication. We implemented PinchKey on the Meta Quest 2 platform and conducted a comprehensive evaluation with 29 users, focusing on key parameters such as authentication accuracy, time, and overall usability. We conducted an offline evaluation using the pre-collected user pinch data. The results revealed high authentication accuracy, with an average area under the curve (AUC) of 99.7% and an average equal error rate (EER) of 1.7% in the initial phase. Based on the above results, we conducted real-time authentication trials on a Meta Quest 2 equipped with PinchKey. The tests emphasized the system's ability to provide efficient and fast authentication, with an average authentication accuracy of 93.8% and a fast average authentication time of 1.65s. PinchKey achieved a commendable median system usability scale (SUS) score of 86.25, which is the highest A rating on the SUS. In addition, we evaluated PinchKey's practicality in real-world applications, including system resilience in the presence of potential attackers.