Energy efficient platform for sobel filter implementation in energy and size constrained systems

Designing a high performance and energy efficient image processing solution for a very limited platform of a small UAV (Unmanned Air Vehicle) is very challenging. We address this issue by conducting a research of low power (under 10 Watt) and small sized (slightly larger than a credit card) embedded platforms with high performance computing capabilities. Sobel filter algorithm used in image processing will be benchmarked using different embedded platforms and frameworks of parallel computing to evaluate energy consumption and image processing performance, thus easing the design selections for a software engineer. The research results show, that Radxa Rock2 platform using Mali T764 GPU appeared to be 6.85x more energy efficient and 3.7x times better performing than Parallella platform using 16 core Epiphany co-processor when computing Sobel filter on 1080p resolution image.