VLSI Implementation of an Artificial Neural Network for a Real Time Detection Problem

Abstract

Real time image processing is an interesting field of application for artificial neural networks as well as for very large scale integrated (VLSI) circuits. This paper shows how the advantage of parallel computing of neural network algorithms can be realized by the potentials of VLSI circuits in order to design a real time detection and tracking system for video image processing. A real time localization and tracking algorithm has been developed for detecting human hands in video-images. Due to the real time aspect, a singlepixel- based classification is aspired, so that a continuous data stream can be processed. Consequently, no storage of full images or parts of them is necessary. The classification, whether a pixel belongs to a hand or to the background, is done by analysing the RGB-values of a single pixel by means of an artificial neural network. To obtain the full processing speed of the neural network a parallel operating hardware solution is realised.