Detecting partial occlusion of humans using snakes and neural networks

Abstract

This paper summarises the development of a computer system designed to detect moving humans in an image or series of images. The system combines the use of active contour models, ‘snakes’, which detect human objects in an image, with a 2 layer feedforward backpropagation neural network, to categorise the detected shape as human, or not. It was found that combining the neural network’s output values with its confidence value provided a means of classifying unseen shapes into ‘human’ and ‘non-human’. Moreover the confidence value can provide a measure of the degree of occlusion of a detected human.