We present BioVision, a bio-mimetics platform based on the human visual system. BioVision relies on the foveal vision principle based on a set of cameras with wide and narrow fields of view. We present in this platform a mechanism for learning visual saliency in an intrinsically motivated fashion. This model of saliency, learned and improved on-the-fly during the robot's exploration provides an efficient tool for localizing relevant objects within their environment. The proposed approach includes two intertwined components. On the one hand, a method for learning and incrementally updating a model of visual saliency from foveal observations. On the other hand, we investigate an autonomous exploration technique to efficiently learn such a saliency model. The proposed exploration, based on the IAC (Intelligent Adaptive Curiosity) algorithm is able to drive the robot's exploration so that samples selected by the robot are likely to improve the current model of saliency. We then demonstrate that such a saliency model learned directly on a robot outperforms several state-of-the-art saliency techniques, and that IAC can drastically decrease the required time for learning a reliable saliency model. We also investigate the behavior of IAC in a non static environment, and how well this algorithm can adapt to changes.