Orb_slam2: How can I process my own video using ORB SLAM2?

[Nish160]

I have taken a video of my desk from mobile phone- Redmi note 4G , Model number-HM NOTE 1LTE.
I have converted the video into a stream of images(.png).
I thought I would have to just save it into folder /cam3/data and write a timestamp file Examples/Monocular/EuRoC_TimeStamps/sample.txt

But in order to make changes in the Examples/Monocular/EuRoC.yaml file I need to know the following:

Camera.fx
Camera.fy
Camera.cx
Camera.cy
Camera.k1
Camera.k2
Camera.p1
Camera.p2
Camera.fps

Color order of the images (0: BGR, 1: RGB. It is ignored if images are grayscale)

Camera.RGB:

My question is how and from where can I know all this information about my mobile phone camera??

The documentation states the following:

"8. Processing your own sequences

You will need to create a settings file with the calibration of your camera. See the settings file provided for the TUM and KITTI datasets for monocular, stereo and RGB-D cameras. We use the calaibration model of OpenCV. See the examples to learn how to create a program that makes use of the ORB-SLAM2 library and how to pass images to the SLAM system. Stereo input must be synchronized and rectified. RGB-D input must be synchronized and depth registered."

But I am still not able to find out the camera calibration parameters.

Any help regarding this is appreciated.

[jamesdeep]

qu wen lei jun ba

[lijuan123]

you should calibrate your camera.

[9friday]

Here is a way, though non-ideal to get started.

1. Save each frame of your video as a separate frame in the PNG format at a resolution of 640x480(according to the TUM1.yaml settings file).

2. Generate a text file like in one of the TUM(or any other) datasets. The file contains the timestamps and filename of each image(frame). You can generate your own timestamps - and I say this again, this is not the ideal way to approach this problem - by taking the first timestamp from one of the text files and add keep that as the timestamp for the first image(frame). Then keep adding 40 ms for the next image(frame) till the last image(i.e. the last frame). You could write a simple program to do this.

3. The images should be saved in a folder named rgb in the main folder(let us name it test_dataset) and the text file should be named rgb.txt and be saved in the test_dataset folder.

4. Then go in the ORB_SLAM2 folder, launch the terminal(I am assuming that you use Ubuntu) and execute the following command:
   ./Examples/Monocular/mono_tum Vocabulary/ORBvoc.txt Examples/Monocular/TUM1.yaml PATH_TO_SEQUENCE_FOLDER
   ,where the PATH_TO_SEQUENCE_FOLDER could be /home/username/test_dataset .

[ujur007]

Did anyone of you tried processing your own sequences with this method suggested by @9friday ?
Does't it needs Depth images also? what about timestamps? there are many of them....

[9friday]

No depth images needed.

On Mon 27 Aug, 2018, 9:35 PM Ujjval Rathod, [e-mail]
wrote:

Did anyone of you tried processing your own sequences with this method
suggested by @9friday https://github.com/9friday ?
Does't it needs Depth images also? what about timestamps? there are many
of them....

—
You are receiving this because you were mentioned.
Reply to this email directly, view it on GitHub
https://github.com/raulmur/ORB_SLAM2/issues/486#issuecomment-416277436,
or mute the thread
https://github.com/notifications/unsubscribe-auth/AkCptYk6A5F5I2sY4fmIx_7NchkF9O5Nks5uVBjWgaJpZM4Q5azs
.

[BW25]

I found that I was able to process video from an iPhone camera.
The advice from 9friday is good, but I found I didn't need to change the resolution of the images.

Also, the time it takes for ORBSLAM2 to process the video is dependent on the timestamp.
If you use intervals of 1 for each timestamp, it will only process 1 frame a second.

[Jackyiyo]

I found that I was able to process video from an iPhone camera.
The advice from 9friday is good, but I found I didn't need to change the resolution of the images.

Also, the time it takes for ORBSLAM2 to process the video is dependent on the timestamp.
If you use intervals of 1 for each timestamp, it will only process 1 frame a second.

Hi BW25
How you process video from iPhone camera? and How you get iPhone calibration variables?

[ujur007]

You can calibrate the camera with some online opencv c++ code.
to process the video you need to create images in rgb folder and time-steps in rgb.txt folder.

[BW25]

I found that I was able to process video from an iPhone camera.
The advice from 9friday is good, but I found I didn't need to change the resolution of the images.
Also, the time it takes for ORBSLAM2 to process the video is dependent on the timestamp.
If you use intervals of 1 for each timestamp, it will only process 1 frame a second.

Hi BW25
How you process video from iPhone camera? and How you get iPhone calibration variables?

My earlier post was incorrect.
ORBSLAM will run without changed resolution, but the feature tracking doesn't quite work right.
I wrote the following bash script to process video into resized .PNGs and create the timestamp file at the same time.
This program assumes the video is in the same folder as the file with the code in it, and is a standard iPhone .MOV video.
It also needs to be run from the terminal.
I found that video shot to be taller vertically doesn't work, so this gives the option to rotate too.
If it doesn't work for you, I hope at least it gives you a framework to work from.

Note that this program relies on ffmpeg, but it came preinstalled with ubunto for me. If not, you can apt-get it.

Hope this helps!

echo
echo "This program processes video for ORBSLAM2"
echo

echo -n "Write the name of the video file in this folder: "
read inputname
echo

if [ ! -e "$inputname" ]
then
echo "File not found. Make sure it is in the same folder as VideoProcessor.sh"
echo
exit 1
fi

echo -n "What would you like the output file to be called: "
read outputname
echo

mkdir -p $outputname/rgb

echo -n "Choose frames per second. Between 10 and 25 is preferable: "
read fps
echo

echo "The video must be wider than it is long"
echo -n "Would you like to rotate it (y/n): "
read rotchoice
echo

if [ $rotchoice = "n" ] || [ $rotchoice = "N" ]
then
ffmpeg -i $inputname -r $fps -vf scale=-1:320 $outputname/rgb/img%04d.png

elif [ $rotchoice = "y" ] || [ $rotchoice = "Y" ]
then
ffmpeg -i $inputname -r $fps -vf scale=320:-1,"transpose=1" $outputname/rgb/img%04d.png

else
    echo "Invalid choice. Choose y/n"
    echo
    exit 1

fi

Counts the number of output files

imgnum=$(ls $outputname/rgb | wc -l)

echo "# colour images" > $outputname/rgb.txt
echo "#file: '$outputname'" >> $outputname/rgb.txt
echo "# timestamp filename" >> $outputname/rgb.txt

Uses bc to calculate timestamp increment to 6 places

No spaces around =

frameTime=$(bc <<< "scale=6; 1.0/$fps")
timestamp=0.000000

for i in $(seq -f "%04g" $imgnum)
do
echo $timestamp rgb/img$i.png >> $outputname/rgb.txt
timestamp=$(bc <<< "scale=6; $timestamp+$frameTime")
done

mv $inputname $outputname

echo
echo "Your files are ready, and have all been put in a single folder."
echo "Please place this folder in ~/Desktop/ORBSLAM2 datasets/our datasets."
echo

[Jackyiyo]

I found that I was able to process video from an iPhone camera.
The advice from 9friday is good, but I found I didn't need to change the resolution of the images.
Also, the time it takes for ORBSLAM2 to process the video is dependent on the timestamp.
If you use intervals of 1 for each timestamp, it will only process 1 frame a second.

Hi BW25
How you process video from iPhone camera? and How you get iPhone calibration variables?

My earlier post was incorrect.
ORBSLAM will run without changed resolution, but the feature tracking doesn't quite work right.
I wrote the following bash script to process video into resized .PNGs and create the timestamp file at the same time.
This program assumes the video is in the same folder as the file with the code in it, and is a standard iPhone .MOV video.
It also needs to be run from the terminal.
I found that video shot to be taller vertically doesn't work, so this gives the option to rotate too.
If it doesn't work for you, I hope at least it gives you a framework to work from.

Hope this helps!

echo
echo "This program processes video for ORBSLAM2"
echo

echo -n "Write the name of the video file in this folder: "
read inputname
echo

if [ ! -e "$inputname" ]
then
echo "File not found. Make sure it is in the same folder as VideoProcessor.sh"
echo
exit 1
fi

echo -n "What would you like the output file to be called: "
read outputname
echo

mkdir -p $outputname/rgb

echo -n "Choose frames per second. Between 10 and 25 is preferable: "
read fps
echo

echo "The video must be wider than it is long"
echo -n "Would you like to rotate it (y/n): "
read rotchoice
echo

if [ $rotchoice = "n" ] || [ $rotchoice = "N" ]
then
ffmpeg -i $inputname -r $fps -vf scale=-1:320 $outputname/rgb/img%04d.png

elif [ $rotchoice = "y" ] || [ $rotchoice = "Y" ]
then
ffmpeg -i $inputname -r $fps -vf scale=320:-1,"transpose=1" $outputname/rgb/img%04d.png

    echo "Invalid choice. Choose y/n"
    echo
    exit 1

fi

Counts the number of output files

imgnum=$(ls $outputname/rgb | wc -l)

echo "# colour images" > $outputname/rgb.txt
echo "#file: '$outputname'" >> $outputname/rgb.txt
echo "# timestamp filename" >> $outputname/rgb.txt

Uses bc to calculate timestamp increment to 6 places

No spaces around =

frameTime=$(bc <<< "scale=6; 1.0/$fps")
timestamp=0.000000

for i in $(seq -f "%04g" $imgnum)
do
echo $timestamp rgb/img$i.png >> $outputname/rgb.txt
timestamp=$(bc <<< "scale=6; $timestamp+$frameTime")
done

mv $inputname $outputname

echo
echo "Your files are ready, and have all been put in a single folder."
echo "Please place this folder in ~/Desktop/ORBSLAM2 datasets/our datasets."
echo

Thanks! I will try it!