Energy efficient object detection on the mobile GP-GPU

Fitsum Assamnew Andargie, Jonathan Rose, Todd M. Austin, Valeria Bertacco

Abstract: Smartphones and tablets now include General Purpose Graphics Processing Units (GP-GPUs) that can be used for computation beyond driving the high-resolution screens. In this paper we present a mobile GP-GPU-based object detection algorithm and system, based on the work by Viola and Jones (which is also used in the OpenCV library). This implementation achieved twofold speed up compared to OpenCV running on the CPU of the same smartphone, and up to 84% energy saving. Interestingly, the new implementation saves energy vs. the CPU even when it executes slower than the OpenCV implementation, because the GPU consumes less power than the CPU, something that is not typical in desktop or laptop systems.