Adaptaci#x00F3;n de modelos de atenci#x00F3;n visual para la localizaci#x00F3;n de sem#x00E1;foros en im#x00E1;genes de situaciones reales de tr#x00E1;nsito

摘要

This project proposes the use of a well-known visual attention model for the construction of a traffic light detection algorithm that works on monocular images. The algorithm design is based on the physiological mechanisms that lead to attention. Its main function is to maximize the probability of the target in the regions of interest (ROIs) found. The algorithm presents a number of contributions that are applied within the computational model of attention. The result is a significant increase of the saliency of traffic lights in images in comparison to the base algorithm. The algorithm has a variety of qualities inherited from the visual attention model. These are: adaptability to find the target in different situations; ability to orient to different target attributes; and the characteristic efficiency of a two-stage processing scheme. This makes the proposed algorithm a viable option to base a traffic lights detection system.