Active vision using an analog VLSI model of selective attention

Detailed processing of sensory information is a computationally demanding t ask. This is especially true for vision, where the amount of information pr ovided by the sensors typically exceeds the processing capacity of the syst em. Rather than attempting to process all the sensory data simultaneously, an effective strategy is to focus on subregions of the input space, shiftin g from one subregion to the other, in a serial fashion. This strategy is co mmonly referred to as selective attention. We present a neuromorphic active -vision system, that implements a saliency-based model of selective attenti on. Visual data is sensed and preprocessed in parallel by a transient image r chip and transmitted to a selective-attention chip. This chip sequentiall y selects the spatial locations of salient regions in the vision sensor's f ield of view. A host computer uses the output of the selective-attention ch ip to drive the motors on which the imager is mounted, and to orient it tow ard the selected regions. The system's design framework is modular and allo ws the integration of multiple sensors and multiple selective-attention chi ps. We present experimental results showing the performance of a two-chip s ystem in response to well-controlled test stimuli and to natural stimuli.