fastHOG – a real-time GPU implementation of HOG
University of Oxford, Department of Engineering Science, Parks Road, Oxford, UK
Department of Engineering Science, Oxford University, Technical Report No. 2310/09, 2009
We introduce a parallel implementation of the histogram of oriented gradients algorithm for object detection. Our implementation uses the GPU and the NVIDIA CUDA framework. We achieve speedups of over 67x from the standard sequential code, using a single video card. Furthermore it supports multiple video cards so speedups of 120x or more can be achieved. This allows us to achieve real-time performance, using the full HOG algorithm for the first time in the literature. All of this is done while keeping compatibility with the standard sequential implementation. Finally our implementation is available online and is open source.