Intrinsic Matrix for simulated camera

Official Python bindings with a focus on reinforcement learning and robotics.
Post Reply
timeforscience
Posts: 3
Joined: Sat Jan 12, 2019 2:11 am

Intrinsic Matrix for simulated camera

Post by timeforscience » Mon Jan 14, 2019 7:03 pm

Hello all! I'm currently trying to do some computer vision tests using the pybullet simulated camera. I can get the image fine, but what I really need for my development is a mapping from world space to pixel space. I don't fully understand the view matrix and projection matrix however. I'm used to an intrinsic matrix where pixel = intrinsic_matrix * transformation_matrix * point_vector. Could someone help guide me just to find a way to create a matrix that will map from world space to pixel space if I have all the other necessary parameters?

I've tried pixel = projection_matrix * view_matrix * point_vector, but that didn't seem to work. Any guidance would be greatly appreciated.

shantythaks
Posts: 1
Joined: Wed Sep 25, 2019 12:50 am

Re: Intrinsic Matrix for simulated camera

Post by shantythaks » Wed Sep 25, 2019 12:52 am

@timeforscience, have you figured out a way to find the matrix to map from world space to pixels? If yes can you please share.

Thanks!

hyyou
Posts: 96
Joined: Wed Mar 16, 2016 10:11 am

Re: Intrinsic Matrix for simulated camera

Post by hyyou » Thu Oct 10, 2019 2:00 am

I studied coordinate conversion

Code: Select all

 model coordinate --> world coordinate --> view coordinate --> projected coordinate.
... from here: https://solarianprogrammer.com/2013/05/ ... iew-model/

If you get projected coordinate, then multiply it by screen resolution, you will get pixel coordinate.

Post Reply