A Task-driven Object-based Attention Model for Robots

Abstract

This paper proposes a task-driven object-based visual attention model for robot applications. It involves five components: pre-attentive object-based segmentation, bottom- up still attention, bottom-up motion attention, top-down object- based biasing, and contour based object representation. The object-based attentional competition operates on the combination of bottom-up saliency map and top-down bias map. This model is applied into two tasks of mobile robots: task-specific moving object detection and still object detection. Experimental results in natural scenes have shown to validate this model even in case of occlusion.