Several industrial and research activities are demanding more robust and efficient image processing systems. Whether they are discrete systems exploiting the capacities of the latest processors and algorithms in the market, or whether they are integrated systems exploiting the advancements in technology: compactness, low power, speed ... This paper describes a real-time vision integrated system for object recognition, based in an APS CMOS matrix. The aim of this work is to demonstrate that we can take some decisions on the fly of the reading process in order to locate and identify objects within the entire image matrix. A neural network circuit will implement the decision subsystem.