Novel signal processing approach for gait based human identification system

Humans can be identified at a distance using gait as a biometric. In this paper, we propose a novel set of features extracted from a walking sequence of a person, which include the varying leg spread, the motion of centroid, the number of pixels on the vertical line through centroid, and the sum of foreground pixels as the dynamic features, and the height and maximum leg spread as the static features for gait identification. These features are very easy to obtain, but contain significant information about the gait of the person. To reduce the size of data, the dynamic features are represented by their respective line spectral pairs. Match scores obtained using Euclidean distance measure on these features are combined with match scores obtained by comparing the horizontal and vertical projection vectors. Nearest neighbor classifier is used and the performance of our recognition algorithm is tested on CASIA A dataset. The results are compared with previous works in this category.