A Hierarchy of Visual Behaviours in an Active Binocular Robot Head

This paper reports current progress into an ongoing investigation of a computational model of hierarchical visual behaviours within an active binocular robot vision system. Our robot vision system is now able to localise endogenously multiple instances of the same object class, while simultaneously maintaining vergence and directing its gaze to attend and recognise these objects within cluttered scenes. Preliminary results show that all identifiable objects defined in our visual search task are correctly identified. An investigation of visual search stability, conducted over 5 trials for 5 different object classes (10 object instances per trial), revealed for object instance fixations: an average fixation variation of ~9.3 (with respect to the object area) and an average spatial correlation of 0.93 between fixation points.