AcousticID

Abstract

Human identification plays an important role in our daily lives. Previous studies have successfully used characteristics such as fingerprints, irises, and facial features for identity recognition. However, these methods require the user being close to the sensing device, which may cause inconvenience to users. In this paper, we present AcousticID, a system that uses fine-grained gait information derived from acoustic signals generated by Commercial Off-The-Shelf devices to identify human beings. We demonstrate the feasibility of gait recognition by analyzing the Doppler effect of various body parts on acoustic signals while walking, and then extract fine-grained gait features that can distinguish different people from both macro and micro dimensions. Similar to an access control system in the home or office, AcousticID is a convenient, low cost, and universal solution. We evaluate AcousticID using experiments with 50 volunteers in an area of 60 m2, and the results show that it can identify different persons with an average accuracy of 96.6%.