Unfortunately, I don't know CUDA yet and didn't implement a GPU accelerated version. I can't comment on its feasibility on mobile GPUs but the paper authors rep

Unfortunately, I don't know CUDA yet and didn't implement a GPU accelerated version. I can't comment on its feasibility on mobile GPUs but the paper authors reported real-time performance on a GTX-480 + i7 quad-core CPU system.