How to get 3d pose from synchronised 2d RGB videos?

[Siba-07]

It was mentioned in these issues, in #797 and #984 but could not find satisfying answers, if anyone figured it out please tell if there is an inbuilt method to get 3d pose from time synchronized existing 2d RGB videos.

[ravijo]

@Siba-07

OpenPose uses camera intrinsic and extrinsic parameters in order to achieve 3D Reconstruction by using the concept of pose triangulation.

Therefore, if you have camera parameters (from which you captured videos) specific to your setup, you can use 3-D Reconstruction Module.

Please refer to the provided documentation above.

Cheers!

[khalidh223]

@Siba-07

OpenPose uses camera intrinsic and extrinsic parameters in order to achieve 3D Reconstruction by using the concept of pose triangulation.

Therefore, if you have camera parameters (from which you captured videos) specific to your setup, you can use 3-D Reconstruction Module.

Please refer to the provided documentation above.

Cheers!

@ravijo I believe the documentation you have specified only outlines how the model would access a single folder of images with the line ./build/examples/openpose/openpose.bin --image_dir input_folder_path/ --3d_views 3 --3d --number_people_max 1 --output_resolution {desired_output_resolution}, but does not describe how the model finds the second or third sequence of images from subsequent cameras, since it is only reading a single folder of images with this line. Same for reading only a single video: ./build/examples/openpose/openpose.bin --video input_folder_path/video.avi --3d_views 3 --3d --number_people_max 1 --output_resolution {desired_output_resolution}

[ravijo]

@khalidh223

Thanks for pointing it out.

The 3-D Reconstruction module provided by OpenPose is supported for FLIR cameras. So if you have FLIR cameras, you can directly follow the instructions. If not then, you need to write CPP code.

All the steps are given in the above documentation. For example, you need to modify 14_synchronous_custom_input.cpp.

I was doing it once with 3 Kinect v2 cameras. It is doable but you need to understand the algorithm and camera parameters as well.

Cheers!

[Siba-07]

@ravijo Isn't the documentation present for multiview camera streams? I couldn't find something similar for already pre-recorded videos.

[khalidh223]

@khalidh223

Thanks for pointing it out.

The 3-D Reconstruction module provided by OpenPose is supported for FLIR cameras. So if you have FLIR cameras, you can directly follow the instructions. If not then, you need to write CPP code.

All the steps are given in the above documentation. For example, you need to modify 14_synchronous_custom_input.cpp.

I was doing it once with 3 Kinect v2 cameras. It is doable but you need to understand the algorithm and camera parameters as well.

Cheers!

@ravijo Is it possible that you have any starter code in which you integrated recorded video/stereo pairs into the 3D reconstruction model? I am struggling to understand this algorithm & cannot find any code that does so.

[hadi1376tm]

i also have the same problem, but i need to capture frames from 3 cameras at same time for 3d model.(not flir)

[hadi1376tm]

@khalidh223 Thanks for pointing it out. The 3-D Reconstruction module provided by OpenPose is supported for FLIR cameras. So if you have FLIR cameras, you can directly follow the instructions. If not then, you need to write CPP code. All the steps are given in the above documentation. For example, you need to modify 14_synchronous_custom_input.cpp. I was doing it once with 3 Kinect v2 cameras. It is doable but you need to understand the algorithm and camera parameters as well. Cheers!

@ravijo Is it possible that you have any starter code in which you integrated recorded video/stereo pairs into the 3D reconstruction model? I am struggling to understand this algorithm & cannot find any code that does so.

HI, did you find the answer?

[stale[bot]]

This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.