Targets Detection Using Multiple Foveas

Target detection enables running a robotic task. However, their limited resources make large amount of data processing harder. Image foveation is an approach that can reduce processing demand by reducing the amount of data to be processed. However, as an important visual stimulli can be attenuated by this reduction, some strategy should be applied in order to keep/recover awareness of it. This work compares gradient descent (potential field), maximum likelihood, multi-lateration, trilateration, and barycentric coordinates to solve this problem in a multiple dynamic and fixed foveas context. Our results demonstrate that the proposed methodology detects the target converging with an average euclidian distance of 51 pixels from the target's center position.