Person Identification and Imposter Detection using Footfall based Biometric System

In this paper, we propose a footfall based biometric system using Gaussian Mixture Model-Universal Background Model (GMM-UBM). Footfall generated ground vibration is used as a biometric modality. We have curated an indigeneous dataset containing 7750 footfall events of twenty subjects. We utilize time and frequency related features extracted from these seismic events. To reduce the feature dimensionality, principal component analysis (PCA) technique is used. We maximize the information gain by augmenting feature set of multiple footfall events to improve the overall accuracy of the system. We demonstrate optimal performance of the system with least number of footsteps per sample while encompassing the maximum number of users. An exhaustive EER analysis is done to evaluate the performance of the system in human identification and imposter detection. We achieve an EER of 7.39% by utilizing 70 principal components of the feature space and by considering 10 footsteps per sample. Experiments have been conducted on the proposed dataset to validate its significance and generalization capability.