Gait-based Person Identification Robust against Speed Variation using CHLAC features and HMMs

概要

論文の詳細を見る
The performance of gait-based person identification is strongly affected by the variations in walking speed. In our previous study, we have proposed a new framework that is robust against speed variation across gait sequences, which combines the Fisher discriminant analysis (FDA)-based cubic higher-order local auto-correlation (CHLAC) features and the statistical framework provided by hidden Markov models (HMMs). The CHLAC features capture the within-phase spatio-temporal characteristics of each walker, while the HMMs identify the person and the phase of each gait even when the walking speed changes nonlinearly. However, since CHLAC features do not have much shape information of a gait phase, it is difficult to identify/segment the walking phase accurately. Therefore in this paper, we not only use CHLAC features to train the HMM, but also utilize principal component analysis (PCA) features that have more shape information of a gait phase in order to have a better gait cycle segmentation/alignment process. We also evaluate our method when the walking speed varied within a gait sequence by manually creating mixed speed variation data within a gait sequence in TokyoTech database. We compared our method with other conventional methods using three other public databases. The proposed method was better than the others when the speed varied across and within a gait sequence.
2010-10-01