thmoa / optflow

Python bindings to optical flow algorithms.

Python Optical Flow

This package provides python bindings to CUDA accelerated optical flow methods.

Methods

Brox Optical Flow

Taken from OpenCV 3.X ¹. OpenCV must be installed on the machine.

To do: Write function to locate OpenCV on the machine without the use of pkg-config.

@inproceedings{brox2004high,
  title={High accuracy optical flow estimation based on a theory for warping},
  author={Brox, Thomas and Bruhn, Andr{\'e}s and Papenberg, Nils and Weickert, Joachim},
  booktitle={European conference on computer vision},
  pages={25--36},
  year={2004},
  organization={Springer}
}

Fast Edge-Preserving PatchMatch for Large Displacement Optical Flow

@inproceedings{bao2014cvpreppm,
  title={Fast Edge-Preserving PatchMatch for Large Displacement Optical Flow},
  author={Bao, Linchao and Yang, Qingxiong and Jin, Hailin},
  booktitle={IEEE Conference on Computer Vision and Pattern Recognition (CVPR)},
  year={2014},
  pages={3534-3541},
  organization={IEEE}
}

Install

git clone https://github.com/linchaobao/EPPM ~/EPPM
ln -s ~/EPPM .
python2 setup.py build_ext -i
python2 demo.py

1: https://docs.opencv.org/3.4.1/d7/d18/classcv_1_1cuda_1_1BroxOpticalFlow.html