A Novel Approach to Behavioral Biometric Authentication for Heterogeneous Actions

Authentication methods based on behavioral characteristics identify individual user behavior patterns and utilize them for user authentication. Compared to traditional methods like passwords and biometrics, this approach offers a more natural and convenient authentication experience. Behavioral biometrics are expected to find applications in areas like ID proofing and surveillance. Authentication methods using accelerometers can directly capture user movements but require users to constantly carry devices, limiting their use in surveillance scenarios. On the other hand, gait recognition based on their walking patterns from general video sequences has gained significant attention. Gait recognition is considered highly distinctive due to subtle variations based on individual physique and physical traits. However, focusing solely on gait presents challenges in static scenarios. We propose a novel method that decomposes the entire skeleton into parts and calculates feature values for each part. Then, it merges the features based on their joint visibility. The proposed method demonstrated higher accuracy compared to a baseline method using full-body features, based on proof-of-concept experiments using the MMAct and CASIA-B datasets.