Energy-Aware High Resolution Image Acquisition via Heterogeneous Image Sensors

Abstract

We present an energy-efficient high resolution image acquisition approach based on a two-tiered system comprising low-cost, low-power, non-actuated, extremely resource-constrained stereo image sensor platforms and more capable but more power-consumptive high resolution imaging platforms with actuation capability. The resource constrained platforms are used to compute 3D object location and subsequently to compute appropriate pan/tilt/zoom settings for the high resolution imaging platforms. The high resolution imaging platforms with actuation capability acquire high resolution images which can be utilized for various recognition purposes. We present a design methodology and system architecture, and evaluate latency and energy tradeoffs in the system. Experimental results show that use of the two-tiered network significantly reduces energy consumption of high resolution image acquisition versus a single-tiered network with minimum loss in detection capability.