Automatic age-group estimation from gait energy images

Abstract

In this paper gait-based method which correlates age with gender information is investigated to estimate human age-group. Estimating age group with one’s gait has gained wide research interest in recent times because the gait can be captured from a distance and does not any co-operation of the person for its acquisition (e.g., CCTV). Gait-based human age group estimation has numerous applications including automatic age-based access to restricted areas, monitoring in public places such as malls for age-dependent customer behavioural analysis. Male and female gaits belonging to same group widely differ and this dependence of gender on age is harnessed by devising a label encoding scheme which codes the age and the corresponding gender information as 8-bit label vector. The prospective gait-based human age estimation is done in three steps. The gait energy image (GEI) obtained from the Gait cycle forms the feature vector and the age and gender information is encoded as a label vector Next, to scale down the feature dimension, Hilbert-Schmidt Independence Criterion (HSIC) is used to learn a low dimension subspace and lastly, using k-nearest neighbour (KNN) classifier, age estimation is carried out and the age information is decoded from the label vector. The experimental results indicate that the proposed method which correlates age information with gender is effective in estimating age-group information from Gait signatures.