Screen coordinates for lower left is (-1.0, -1.0)
Screen coordinates for for upper right is (1.0, 1.0)

So you have your (mouse_x, mouse_y) in screen coordinates...

Now you get your world coordinates for  lower left and upper right portion of the screen... how? I did my mouse control a different way but I would try base.camLens.getFilmSize() or other functions from the global object base.camLens (type is Lens*). From this object and knowing the current camera position, base.camera.getPos(), you should be able to compute the world coordinates of the screen corners. Then you do the ratio math from there. Too bad there isn't a corresponding get function to base.camLens.setFrustumFromCorners() in Panda3D. You have to do the calculation yourself.

After you get the coordinate conversion done, shoot a collision ray from the point in the same direction the camera is facing. If you get a collision, you hit a target.

Easier way to do this homework is to parent a collision ray to the camera, facing the same direction... wherever camera ray points, collision ray will follow (see roaming ralph example). Then use mouse to control heading and pitch of camera (see  below).

could you post a picture of whatcha got so far? for any perspective projection i don't see how you could avoid using collision rays. it would help to see a pic.

ha ha ha panda3d has functions built in just for this. and i see you started...

http://www.panda3d.org/wiki/index.php/Clicking_on_3D_Objects

you should try it. that

pickerRay.setFromLens(base.camNode, mpos.getX(), mpos.getY())

sets the collision ray from the mouse position automatically for you, rather than having to calculate it ourselves from the lens parameters like we were trying to do.

just create the ray in the constructor, set a mouse1 event, and in the mouse1 event set the ray using the setFromLens function they talk about. then traverse the collision objects and check for a hit. the reason they use getTag and findNetTag is because one model loaded multiple times in the scene will yeild collision entries with the same name (the name of the model), and you can't tell them apart, so you have to tag each model with a unique name when you create them (I used bandit1, bandit2, bandit3 and the model name is bandit).