Automatic Extraction of Spatio-Temporal Gait Features for Age Group Classification

Automatic age estimation involves the automatic labeling of a human with a precise age or age group based on physical attributes. It is important in human–computer interaction, surveillance, security and law enforcement. Most automatic age estimation techniques rely on face images. A major limitation of face-based age estimation is that performance depends on face image quality, whereas most surveillance cameras produce low-resolution images. It is also challenging to capture face images unobtrusively. More recently, it has been discovered that human gait also contains age discriminant features. Unlike face, gait features can be obtained unobtrusively and from a distance. However, many gait-based age estimation techniques have large estimation errors due to large variations in age and because the models do not consider different age groups. Also, gait feature extraction still suffers from a lot of variation. Results obtained from gait is affected by factors including view angles, apparel and walking speed. We propose a framework for automatic gait-based age group estimation. Experiments were performed using six extracted biological and dynamic gait features, and an overall accuracy of 96% was achieved.