Incorrect wide angle camera lens flare occlusion after camera movement

[mogumbo]

Environment

• Ubuntu 20.04
• gazebo11 d358400

Description

• Expected behavior: After moving a robot with a wide angle camera and then triggering an image, lens flare should be occluded or not occluded as expected.
• Actual behavior: Sometimes lens flare is not occluded by objects, and sometimes lens flare does not appear when there is no object to occlude it.

Steps to reproduce

After moving your robot this call to occlusionScale will sometimes give an incorrect result: https://github.com/gazebosim/gazebo-classic/blob/gazebo11/gazebo/rendering/LensFlare.cc#L314

If you inspect the view matrix just before that call to occlusionScale, it will not be up to date. Move your robot, trigger an image, and then trigger a second image. You should see that the view matrix changes between image triggers. Here's some code for inspecting the view matrix:

            const Ogre::Matrix4& m = this->dataPtr->wideAngleDummyCamera->OgreCamera()->getViewMatrix();
            std::cerr << m[0][0] << "  " << m[0][1] << "  " << m[0][2] << "  " << m[0][3] << std::endl;
            std::cerr << m[1][0] << "  " << m[1][1] << "  " << m[1][2] << "  " << m[1][3] << std::endl;
            std::cerr << m[2][0] << "  " << m[2][1] << "  " << m[2][2] << "  " << m[2][3] << std::endl;
            std::cerr << m[3][0] << "  " << m[3][1] << "  " << m[3][2] << "  " << m[3][3] << std::endl;

Digging a little deeper, you can see that in Scene::FirstContact https://github.com/gazebosim/gazebo-classic/blob/gazebo11/gazebo/rendering/Scene.cc#L1217 the origin and dir used to compute occlusions are incorrect during the first image trigger because the view matrix is incorrect.

Output

[scpeters]

I added an additional model with a pair of wide-angle cameras updated at 1 Hz to a test world in a43e4a3 and was able to reproduce the issue:

wide-angle-lens-flare-occlusion-lag-720.mov

[scpeters]

there is a fix in review in #3325