The system presented in this paper consists of an image acquisition module from a video camera, an image processing module, which performs conversion to monochrome format, and a neural network that takes the acquired images and performs the extraction of patterns from them. The system proposed by the authors is used for recognizing the orientation of objects in real time by analyzing images from a common video camera. The acquisition module is based on three high-speed AD converters. The remaining components: the digital image processor and the neural network used to classify the shapes of objects are both integrated into a FPGA circuit. All these ensure integration in one chip and a high speed response - typical requirements for a real-time system.