Footstep-Identification System Based on Walking Interval
Xuan Wang, Tengfei Yang, Yao Yu, Ruixin Zhang, Fangxia Guo · IEEE Intelligent Systems · 2015
Footsteps, as a main kind of behavioral trait, are a universally available signal, but constructing an identity verification system based on them remains a challenging problem: footsteps not only reflect a person's physiological basis but also depend on the person's psychological makeup, footwear, and floor. This article describes a novel footstep-identification system. To eliminate footwear and floor variations as limiting factors, the footstep duration and interval times are extracted from footsteps, and a timing vector is obtained as a feature. To smooth instability in footsteps, the authors developed a novel pattern-recognition method, in which the training procedure can be split into several parallel subprocedures, with each subprocedure only considering one class sample. It can be periodically retrained using several of the user's most recent successful identification footsteps. Theoretical and experimental results show this system is relatively robust to the variations of footwear, floor, and the examinee's psychological makeup, and yields a better classification performance compared with the existing methods.