Accelerating Computer Vision on mobile embedded platforms
Rahul Kumar Singh, Lakmal Ranasinghe · 2016
Running Computer Vision (CV) algorithms on a mobile embedded platform includes multiple challenges due to high computation requirement. Past few years have witnessed remarkable development in the computational capabilities and applications of hardware accelerators such as GPU, DSP, FPGA and multi-core CPUs in computer vision arena. The computational capabilities of these hardware accelerators are improving day-by-day. Nevertheless, this has also concluded in a significant increase in their power consumption. In this paper, we have surveyed several methods that can optimise the performance of CV algorithm on a mobile embedded platform, also demonstrated that use of GPGPU[11] can significantly improve latency compare to other CPU only optimisation techniques. This demonstrates a valid approach for implementing CV algorithms on GPGPU based parallel computing embedded platforms.