Detecção de estímulo visual usando múltiplas fóveas

Abstract

The multifoveation technique allows to add several focus in the image, which can be explored as points of visual attention in contexts of object detection, identification and/or recognition. However, the use of multifoveation technique requires knowledge of the position of the visual stimuli. In this work we propose a new approach to detect visual stimuli using the structure of multiple foveas. For this, we use mathematical strategies adapted to the context of computer vision, which consider the distribution of the foveas to estimate the localization of the visual stimuli in the image. The mathematical strategies adopted were the gradient descent (potential field), maximum likelihood, multilateration, trilateration and barycentric coordinates. The results show that the algorithms converge to the position of the visual stimulus, with the exception of the intersection of potential locations algorithm due to sensitivity to local minimums. In addition, algorithms that use the potential field to require more processing time and computational resources compared to other strategies. However, it is possible to affirm that three foveas are enough to estimate the position of the visual stimulus in the image using the trilateration and barycentric coordinates algorithms. We conclude that the multifoveation associated with mathematical strategies can be applied in visual detection and converges with at least three foveas.