VirtualGimbal

The VirtualGimbal is an electronic stabilize device for videos that were taken by a hand-held camera such as a DSLR. For more information, see [PetaPixel] ( https://petapixel.com/2016/08/11/sd-card-built-gyro-sensor-stabilize-shots/ "PetaPixel"). The VirtualGimbal consists of an SD card and post-processing software. This repository releases the post-processing software, a SD card circuit and a SD card firmware.

Demo

https://youtu.be/NwUnrnFLjD4

Features

1.Post-processing video stabilization software on PC.
2.4K video available.
3.Stabilization completely based on captured angular velocity data.

Requirement

Ubuntu 18.04 or later FFMPEG, OpenCV3, Eigen, boost and some python libraries.

Usage

See post_processing_software/README.md
Download Example Video, put in ~/vgdataset. Download Angular Velocity data and put it in virtualGimbal/post_processing_software/build/records. A camera calibration file also required to run a demo below so download Calibration file then put it in virtualGimbal/post_processing_software/build/camera_descriptions

Demo:

./pixelwise_stabilizer -i ~/vgdataset/C0003.MP4 -c ILCE-6500 -l SEL1670Z -j records/2019-04-03_07.27.36.json -z 1.3 -o

Here, i option is an input video file. c is a camera name that is calibrated using a calibrator. l is a lens name. j is a json file that include an angular velocity record of a gyro sensor. z is a zooming ratio. Recommended zooming ratio is 1.3 . o option outputs stabilized video.
Additional option n is available to disable preview window. This option speed ups the processing.

Camera calibration is required before you stabilize your own video. See a detail information to do it.

Install dependencies

Install system dependencies:

sudo apt-get install cmake make git

Install Git:

sudo apt install git-all

Install Python Dev:

sudo apt-get install python3-dev python3-pip python3-tk python3-lxml python3-six

Install Numpy & Scipy:

sudo apt-get install python3-numpy python3-scipy

Install Matplot Library:

sudo apt-get install python3-matplotlib

Install OpenCL: See detail information

sudo apt install ocl-icd-libopencl1 opencl-headers clinfo ocl-icd-opencl-dev  

# OPTION: Install intel conpute-runtime with a command below if you use intel HD graphics.
# If you use NVIDIA gpu, it is not required. On the other hand, CUDA is required.
sudo apt-get install intel-opencl-icd

Install ffmpeg:

sudo apt install yasm pkg-config  
cd  
git clone  git://source.ffmpeg.org/ffmpeg.git  
cd ffmpeg
git checkout -b tag refs/tags/n4.4
./configure --enable-gpl --enable-x86asm --enable-opencl --enable-ffmpeg --enable-pic --enable-shared --disable-static --enable-swscale
make all -j4  
sudo make install  
sudo ldconfig

Install OpenCV

cd  
sudo apt-get install build-essential module-assistant libgtk-3-dev  
sudo m-a prepare  
git clone https://github.com/opencv/opencv.git
cd opencv
git checkout refs/tags/4.5.3
cd opencv/  

mkdir build  
cd build  
cmake .. -DWITH_TBB=ON -DWITH_OPENGL=ON -DWITH_VTK=ON -DWITH_GTK=ON  -DWITH_FFMPEG=ON -DCMAKE_BUILD_TYPE=Release    
make -j4  
sudo make install  

sudo sh -c "echo /usr/local/lib >> /etc/ld.so.conf.d/opencv.conf"  
sudo sh -c "echo PKG_CONFIG_PATH=\\\$PKG_CONFIG_PATH:/usr/local/lib/pkgconfig >> /etc/bash.bashrc"  
sudo sh -c "echo export PKG_CONFIG_PATH >> /etc/bash.bashrc"

Install Eigen

sudo apt install libeigen3-dev

Build

cd
git clone https://github.com/yossato/virtualGimbal.git  
cd virtualGimbal  
mkdir build  
cd build  
cmake ../post_processing_software/ -DCMAKE_BUILD_TYPE=Release  
make -j4

To start up faster

I added a program angular_velocity_estimator that generates a JSON file. The file will be saved in same directory of video file after execution. If the file exists, the pixelwise_stabilizer starts up faster. To run the program, execute a command like below.

./angular_velocity_estimator ~/vgdataset/C0003.MP4

This command generates C0003.json .

The file contains estimated angular velocity from the video that is used for synchronizing between video and angular velocity measured by the gyro sensor.

yossato / virtualGimbal

readme