A system that applies a Neural Network to a video surveillancesystem that consists of a pre—processing unit that extracts high level information from images and intruduces it in the Neural Network that can learn in operational conditions while under the supervision of an unskilled operator.
ABSTRACT It is presented a system that applies a Neural Network to a video surveillancesystem. It consists of a pre—processing unit that extracts high level informationfrom images and intruduces it in the Neural Network. This system can learn in operational conditions while under the supervision of an unskilled operator. It was used the error backpropagation learning algorithm in a multilayer perceptronstructure. The results obtained show that the system performs well and with a high degree of efficiency. INTRODUCTION In the last two years we have been developing an image processing system in order to make easier the task of surveillance [1]. In the present state the system hasimplemented in hardware the capacity of signalizing ,in real time, the movement ofany object (that we shall call intruder). It can also accumulate in the same image the signalization made in past images. To this charcteristic it has been noticed the necessity of joining one other: the possibility of an unskilled operator teaches the system how to diferenciatte between"friendly" intruders and "unfriendly" ones. From the several possible aproaches thechoice has fallen in a Neural Network because it's ease to teach. The chosen Neural