Invited Talk 1: Gait-based Age Analysis

In this keynote talk, we will describe a series of our work on gait-based age analysis. First, we will introduce a former version of our gait database for gait-based age estimation and its initial performance evaluation using a standard gait features such as gait energy image (GEI) and frequency-domain feature as well as a standard regression method, i.e., Gaussian process regression [1]. We will then introduce our recent, the world largest gait databases for age estimation [3] which were collected in conjunction with long-term exhibition of intuitive video-based gait analysis demonstration in a science museum [2]. We also introduce performance evaluation results of several benchmark methods of gait-based age estimation on it. Thereafter, we introduce two recent approaches to gait-based age estimation. One is a hierarchical approach, where a gait sample is projected into a low-dimensional embedding space, then classified into one of age groups by a directed acyclic graph support vector machine, and finally fed into an age group-dependent support vector regressor to estimate its age [4]. The other one is based on a state-of-the-art deep learning architecture, i.e., DenseNet [5]. In addition to the above-mentioned gait-based age estimation, we also construct an age progression/regression model based on mean GEI of each age group and free-form deformation between the mean GEIs of adjacent age groups [6], which has potential applications to gait recognition robust against time elapse and physical gait age estimation.