In this article, we present a person identification system for Internet-of-Things-based smart home applications that utilizes footstep induced structural vibrations as biometric modality. Footfall events are generated by the rhythmic contact of the heel and toe on the floor while walking. In the case of such biometric systems, there is no disturbance of the natural movement of the individuals and, thus, they provide an advantage over the existing systems that deal with human intervention. Another key advantage is that footfall signals are not subjected to spoofing attacks as they are nearly impossible to mimic unlike other biometric traits (fingerprint, face, and voice). We propose a 3-layer computing architecture, for the decentralized implementation of the biometric system. We also propose a basis pursuit-based data compression technique (DS8BP) to reduce power and bandwidth for the wireless transmission of footfall events. DS8BP achieves a compression ratio of 108 and increases the scalability of the system. We performed extensive experimentation to evaluate the proposed biometric system using indigenous databases containing 100,000+ footfall events of eight individuals collected in three types of surfaces (concrete tile floor, carpet floor, and wooden floor). The proposed system achieves a prediction accuracy of 93%, 98%, and 96% in three surface types when features from seven footsteps are considered.