The project function in the bundle adjustment g2o example seems to be using the current camera pose estimate to transform the 3d landmark point. However, the 3d landmark obtained from the Washington dataset is already in the world coordinates. Shouldn't this step multiply with the inverse of the camera pose estimate, so that the 3d landmark is transformed from world coordinates into camera coordinate system, before applying the perspective projection?
I would also appreciate if someone could explain how we obtained the pixel coordinates in the project function, since it does not look like the standard camera projection equation.
Hi,
The project function in the bundle adjustment g2o example seems to be using the current camera pose estimate to transform the 3d landmark point. However, the 3d landmark obtained from the Washington dataset is already in the world coordinates. Shouldn't this step multiply with the inverse of the camera pose estimate, so that the 3d landmark is transformed from world coordinates into camera coordinate system, before applying the perspective projection?
https://github.com/gaoxiang12/slambook2/blob/f522131fadae7e7ec1a09f18873bc65369cdd073/ch9/bundle_adjustment_g2o.cpp#L65
I would also appreciate if someone could explain how we obtained the pixel coordinates in the project function, since it does not look like the standard camera projection equation.