Sample Size in Floor Sensor-based Gait Recognition for Smart Home and Access Control Scenarios

Floor sensor-based gait recognition is an emerging biometric technology that can be used in a variety of scenarios ranging from access control to personalized recognition for smart home systems. These distinct scenarios can differ based on: (1) the number of users in the system, N-user, and (2) the number of training footstep samples per user made available during the enrollment process, N-step. In this study, the effect of these two parameters on the accuracy of state-of-the-art machine learning (ML) and deep learning (DL) models was investigated. For the smart home scenario (small N-user, large N-step), a best person verification performance of 99.77% accuracy was found using a lightweight convolutional neural network (CNN) model with spatial image features. For the access control scenario (large N-user, small N-step), on the other hand, a highest accuracy of 90.70% was found using a shallow k-nearest neighbor classifier with spatiotemporal information, and not end-to-end CNNs or transfer learning based on a pre-trained ResNet-50. Finally, a learning curve analysis is conducted to inform how many training footsteps are needed for both ML and DL approaches to person verification based on different numbers of users.