jadarve / optical-flow-filter Goto Github PK

A real time optical flow algorithm implemented on GPU

License: BSD 3-Clause "New" or "Revised" License

C++ 35.17% Python 1.96% CMake 0.10% Cuda 2.51% Jupyter Notebook 59.93% C 0.02% MATLAB 0.32%

robotics gpgpu cuda computer-vision gpu real-time

optical-flow-filter's Introduction

IMPORTANT:

Please consider using https://github.com/jadarve/lluvia instead of this repo. Lluvia includes an implementation of this algorithm running using the Vulkan API instead of CUDA. Checkout this demo: https://www.youtube.com/watch?v=mRZ6YdWb8fE.

Optical-flow-filter

A real time optical flow algorithm implemented on GPU.

@Article{2016_Adarve_RAL,
  Title = {A Filter Formulation for Computing Real Time Optical Flow},
  Author = {{Juan David} Adarve and Robert Mahony},
  Journal = {Robotics and Automation Letters},
  Year = {2016}
}

Build and Installation

Dependencies

CMake 2.8.11 or higher.
Cuda 7.5 or higher.
GCC 4.8.
Visual Studio 2013 (Windows only).

Build (Linux)

git clone https://github.com/jadarve/optical-flow-filter.git
cd optical-flow-filter
mkdir build
cd build
cmake ..
make
sudo make install

The library and header files will be installed at /usr/local/lib and /usr/local/include respectively.

Build (Windows)

For x86_64

mkdir build64 & cd build64
cmake -G "Visual Studio 12 2013 Win64" ..
cmake --build . --config Release

For x86

mkdir build & cd build
cmake -G "Visual Studio 12 2013"
cmake --build . --config Release

Python Wrappers

A python package with wrappers to the C++ library is available at optical-flow-filter/python/ folder. The wrappers have been developed and build using Cython 0.23.4.

cd optical-flow-filter/python/
python setup.py build
sudo python setup.py install

See notebooks/ folder for usage examples.

Demo Applications

flowWebCam

This demo computes optical flow from a webcam. It uses OpenCV to access the camera video stream and to display the computed flow. The instructions to build the demo are the following:

cd optical-flow-filter/demos/flowWebCam
mkdir build
cd build
cmake ..
make
./flowWebCam

highSpeedDemo

This demo interfaces a Basler camera, in our case an acA2000-165um, with the GPU optical flow algorithm, and displays the color encoded flow.

cd optical-flow-filter/demos/highSpeedDemo
mkdir build
cd build
cmake ..
make

To run the application, it is necessary to specify the camera properties file, as follows

./highSpeedDemo -c ../acA2000-165um_binSkip.pfs

Other optional arguments are:

./highSpeedDemo -h

-h, --help             Displays this help.
-v, --version          Displays version information.
-c, --config <file>    Camera configuration file.
-l, --levels <int>     Flow filter pyramid levels (default 2).
-r, --rate <int>       Camera frame rate (default 300).
-m, --maxflow <float>  Maximum optical flow (default 4.0).

optical-flow-filter's People

Contributors

Stargazers

Watchers

optical-flow-filter's Issues

Add Windows compatibility.

Modify the code and CMakeLists.txt files so that a .dll can be created and used on Windows applications.

Check https://msdn.microsoft.com/en-AU/library/ms235636.aspx for reference on dll export/import.

Assign to branch windows_dll

Add functionality for predicting images given rotational optical flow

Given some optical flow w and camera intrinsic parameters, generate the rotational optical flow and use it for predicting the image at some point in time.

Use Lax-Wendroff scheme for numerical solution of the prediction.

How to calculate the backward flow?

Hi,
Really appreciate your sharing. I am now implementing this algorithm to calculate both the forward optical flow ((t-1)->t) and the backward optical flow (t->(t-1)). However, the code
"###filter.loadImage(hostImageGray);
###filter.compute();
###filter.downloadFlow(flowHostWrapper);"
I use can only calculate the forward optical flow ((t-1)->t). May I wonder how to calculate the backward optical flow based on your shared code?

Best,
Jiangtian

Optical flow always equals zero?

Hi Juan,

Thanks for sharing your code. I'm trying to incorporate your optical flow calculations in my project and have followed your webcam demo to calculate flow using OpenCV. I'm acquiring camera image frames from a pre-recorded video for my processing and always get a flow vector of zero magnitude. I also get the following message on my console:

DeltaFlowFilter::compute(): fisrt load
FlowFilter::compute(): fisrt load
DeltaFlowFilter::compute(): fisrt load
DeltaFlowFilter::compute(): fisrt load
FlowFilter::compute(): fisrt load

This message repeats itself throughout video processing. Do you have any advice? I get the feeling that only a single optical flow is being calculated for the initial frame and state propagation isn't taking place. Thanks for your help!

A problem during installation in python modules.

I was trying install the python module, but after I input "python setup.py install" I got an error: Unable to find vcvarsall.bat(seen in the picture). So how can I fix it. Any help will be appreciate.

where is the "flow magnitude" frame?

How do I get the "flow magnitude(pixel)" as I could see in this following video?
300 Hz Real-time Optical Flow. ANU campus

error: libflowfilter_gpu.so

when I run ./flowWebCam, errors occured.
./flowWebCam: error while loading shared libraries: libflowfilter_gpu.so: cannot open shared object file: No such file or directory
Any help?
Best,
Joy Yang

Question about paper's equation (17).

Hi Juan,

Sorry, can I ask some question about the paper here?
Why Equation (17) uses ^h\Delta\Phi rather than ^h\Phi ?
I have not traced your code yet but this has already perplexed me for a long time.

Your paper is great, and code work fine to me! Thanks for your help!

Error occuring in setup.py ,due to which ipython notebook is notworking

1.The error which is occurring again and again in the command sudo python setup.py install is
Traceback (most recent call last):

File "setup.py", line 17, in <module>
   from Cython.Build import cythonize
ImportError: No module named Cython.Build

2.Also the python setup.py build command is not installing or building actually. The command just gets executed within 2-3 seconds. So i guess it's not doing anything in actual. When the command was executed for the first time two warnings were printed several times ,

 warning: command line option ‘-Wstrict-prototypes’ is valid for C/ObjC but not for C++
warning: #warning "Using deprecated NumPy API, disable it by " "#defining NPY_NO_DEPRECATED_API NPY_1_7_API_VERSION" [-Wcpp]
 #warning "Using deprecated NumPy API, disable it by " \

and after executing that command again
It is printing following lines

running build
running build_py
running build_ext

The cuda version

nvcc: NVIDIA (R) Cuda compiler driver
Copyright (c) 2005-2017 NVIDIA Corporation
Built on Fri_Sep__1_21:08:03_CDT_2017
Cuda compilation tools, release 9.0, V9.0.176

and the os is

Distributor ID:	Ubuntu
Description:	Ubuntu 16.04.4 LTS
Release:	16.04
Codename:	xenial

So how to fix this ?
And is there any other method to achieve the same results for optical-flow-filter?

A real time optical flow algorithm implemented on GPU is it CPU compatible?

Can I run this code on laptop using core i7 CPU?
I am sorry if my question is trivial, but is the code implementation has something special to work on GPU only?