I am looking at the following link:
https://docs.omniverse.nvidia.com/app_isaacsim/prod_materials-and-rendering/103/cameras.html
Are the properties nominal width/height, offset center X/Y referring to the camera intrinsic values fx,fy,cx,cy?
Is there a way to insert camera intrinsic values and distortion coefficients for popular opencv camera models such as plumb_bob?
Thanks.
Hi Naruto Uzumaki,
We are working on adding support for popular camera models, like the OpenCV plump_bob. In the current release of Isaac Sim, this model is not supported directly by the RTX renderer. But, the following workaround could be used, to allow fitting an existing OpenCV distortion model into the supported by Isaac Sim fTheta model.
Please, refer to the following example. Please, don’t hesitate to contact us, if you need clarification or have further questions.
%pip install opencv-python scipy
import cv2, numpy as np
from scipy.optimize import curve_fit
# Given the OpenCV camera matrix and distortion coefficients calculate the Isaac Sim fisheyePolynomial distortion
width, height = 1920, 1200
camera_matrix = [[958.8, 0.0, 957.8], [0.0, 956.7, 589.5], [0.0, 0.0, 1.0]]
distortion_coefficients = [0.14, -0.03, -0.0002, -0.00003, 0.009, 0.5, -0.07, 0.017]
((fx,_,cx),(_,fy,cy),(_,_,_)) = camera_matrix
def DistortPoint(x, y, camera_matrix, distortion_coefficients):
((fx,_,cx),(_,fy,cy),(_,_,_)) = camera_matrix
pt_x, pt_y, pt_z = (x-cx)/fx, (y-cy)/fy, np.full(x.shape, 1.0)
points3d = np.stack((pt_x, pt_y, pt_z), axis = -1)
rvecs, tvecs = np.array([0.0,0.0,0.0]), np.array([0.0,0.0,0.0])
cameraMatrix, distCoeffs = np.array(camera_matrix), np.array(distortion_coefficients)
points, jac = cv2.projectPoints(points3d, rvecs, tvecs, cameraMatrix, distCoeffs)
return np.array([points[:,0,0], points[:,0,1]])
def Theta(x, y, camera_matrix):
((fx,_,cx),(_,fy,cy),(_,_,_)) = camera_matrix
pt_x, pt_y, pt_z = (x-cx)/fx, (y-cy)/fy, 1.0
r2 = pt_x * pt_x + pt_y * pt_y
theta = np.arctan2(np.sqrt(r2), 1.0)
return theta
def poly4_00(theta, b, c, d, e):
return b*theta + c*np.power(theta,2) + d*np.power(theta,3) + e*np.power(theta, 4)
# Fit the fTheta model for the points on the diagonal.
X = np.linspace(cx,width, width)
Y = np.linspace(cy,height, width)
theta = Theta(X, Y, camera_matrix)
r = np.linalg.norm(DistortPoint(X, Y, camera_matrix, distortion_coefficients) - np.array([[cx], [cy]]), axis=0)
order4_00_coeffs, _ = curve_fit(poly4_00, r, theta)
# The coefficient 'a' of the Ftheta model is set to zero, so that angle 0 is at pixel distance (aka radius) 0
Ftheta_A = [0.0] + list(order4_00_coeffs)
Dfov = np.rad2deg(2*poly4_00(np.linalg.norm([height/2,width/2]) , *order4_00_coeffs))
print('hfov {}'.format(np.rad2deg(2*poly4_00(width/2, *order4_00_coeffs)))) # assumes fisheye center is image center
print('vfov {}'.format(np.rad2deg(2*poly4_00(height/2, *order4_00_coeffs))))
print('dfov {}'.format(np.rad2deg(2*poly4_00(np.linalg.norm([height/2,width/2]) , *order4_00_coeffs))))
print('Ftheta_A:', Ftheta_A)
# To set the distortion model in the Isaac Sim, please refer to:
# https://docs.omniverse.nvidia.com/app_isaacsim/app_isaacsim/isaac_sim_sensors_camera.html
# And set the parameters either using the GUI Example or define the polynomial by adding to the Python Example
#
# Pixel_size = 3 # in µm
# camera.set_resolution((width, height))
# camera.set_focal_length((fx * Pixel_size * 1e-3 + fy * Pixel_size * 1e-3) / 2) # in mm
# camera.set_horizontal_aperture(Pixel_size * 1e-3 * width) # in mm
# camera.set_vertical_aperture(Pixel_size * 1e-3 * height) # in mm
#
# camera.set_projection_mode("perspective")
# camera.set_projection_type("fisheyePolynomial")
# camera.set_stereo_role("left")
#
# camera.set_fisheye_polynomial_properties(
# nominal_width=width, nominal_height=height,
# optical_centre_x=cx, optical_centre_y=cy,
# max_fov = Dfov, polynomial = Ftheta_A)
Thank you for this example. To get Isaac Sim to do the lens distortion while rendering instead of postprocessing, what are the fTheta coefficients k0-k5 and Projection Type to put into the camera parameters in Isaac Sim, given the OpenCV camera_matrix and distortion_coefficients? Thanks!
Hi Eric. Welcome!
The code sample above explains onto how to calculate the the fTheta coefficients k0-k4 of the fisheye polynomial model from the OpenCV rational polynomial model. And the comment at the bottom of that code provides a reference onto how to set these coefficients for the Isaac Sim camera sensor.
I’d recommend first, follow the tutorial in the this code sample that explains how to set the camera parameters.
And running it either in a notebook or as a script. And then adding the code above, to calculate the distortion parameters based on your OpenCV model. We’ll provide simplified support for the OpenCV rational polynomial model [extrinsics/intrinsics], but for now the tutorial and the sample above are likely the best starting point.
Thanks for the clarification, Dmitry! The crucial thing you pointed out is to calibrate using the fisheye polynomial model from the OpenCV rational polynomial model, rather than the standard radial-tangential distortion model (AKA Plumb Bob or Brown-Conrady or Original OpenCV model). Although the lens is not actually fisheye (i.e., FOV < 120°), this fTheta conversion approximates the radial distortion without tangential distortion, so it’s pretty close!
Looking forward to using the OpenCV plump_bob model in Isaac Sim when that’s ready, but this gets us by for now.
Thanks!
Eric
Thank you very much @dchichkov .
This example helps.
One quick question: Will I also be able to use the camera model above and publish it to ROS? Would that need to happen only using the ActionGraph?
could you please help how to set the camera and its intrinsics to be D435 from Intel RealSense?
how can we insert camera intrinsic values and distortion coefficients for the fisheyePolynomial camera model in omniverse?
for anyone else looking for information on this, take a look at the new doc:
https://docs.omniverse.nvidia.com/isaacsim/latest/features/sensors_simulation/isaac_sim_sensors_camera.html?highlight=rational#calibrated-camera-sensors