0

I am working on human pose estimation work. I am able to generate 2d coordinates of different joints of a person in an image. But I need 3d coordinates to solve the purpose of my project.

Is there any library or code available to generate 3d coordinates of joints ?

Please help.

User4680
  • 1
  • 1
  • Hi, welcome to StackOverflow! You should add more details such as sample of your code and your execution log as it will help us help you. You can also take the [Tour](https://stackoverflow.com/tour), it will help you understand how to properly ask a question :) – vmf91 Apr 24 '21 at 23:44
  • You could try something like this: https://github.com/facebookresearch/VideoPose3D. 3d pose estimation from monocular images is obviously not as accurate as using a depth sensor or multiple cameras, but you can get plausible looking 3d skeletons out. – Will Apr 28 '21 at 22:37

1 Answers1

1

for 3d coordinates on pose estimation there is a limit for you. you cant get 3d pose with only one camera (monocular). you have 2 way to estimate those :

  1. use RGBD ( red, green, blue and depth) cameras like Kinect

  2. or use stereo vision with using at least two camera.

for RGBD opencv contrib has a library for that.

but if you want to use stereo vision you have some steps:

1.Get camera calibration parameters

for calibration you can follow this.

2.then you should get undistorted of your points with using calibration parameters.

3.then you should get projection matrix of your both cameras.

4.at last, you can use opencv triangulation for getting 3D coordinates.

for more info about each step, you can search about stereo vision, camera calibration, triangulation and etc.

Amir Karami
  • 181
  • 7