Human shape recognition from snakes using neural networks

Abstract

This paper documents experiments which have been carried out with several neural network systems designed to categorise pedestrian shapes from non-pedestrian shapes. Active Contour models (‘Snakes’) [1] have been used to obtain contours of pedestrians as they move around the visual field. Neural networks have then been trained on representations of these relaxed snakes. The neural network systems developed can successfully discriminate these contours based upon whether they are ‘pedestrian’ in shape or not. Results are presented along with a discussion of some of the system’s possible applications.