I need to implement what amounts to a "Google Maps" style zoom effect in WebGL. Specifically I have a simple 2-dimensional scene that is always perpendicular to the camera. When a user clicks on the scene, the camera should zoom to a location that is over the click, but closer to the 2-dimensional scene.
For example see this jsfiddle, that implements the scene but no zooming:
If you have a WebGL enabled browser, you should see a triangle and a square (2-dimensional) rendered at -7 on the Z axis. I have put in a placeholder handleMouseUp() event handler that logs any click events, but I'm a little lost as to how to translate the coordinates given by the click event into a new location for the camera (or I guess equivalently a new view matrix).
(The jsfiddle is based on tutorial 1 from learningwebgl.com and uses the glMatrix http://code.google.com/p/glmatrix/ library for matrix operations. Keep in mind that this is WebGL, which is similar to OpenGL ES, and has no access to glu* functions.)
I've written something in this jsfiddle that should help you.
https://jsfiddle.net/pliepa/Lfroxj02/01
(or https://codepen.io/brainjam/pen/gBZyGm)
Just click on the WebGL window to zoom in to where the mouse is pointing.
The basic idea is that a point in the WebGL window is obtained by projecting it from 3-space using the projection matrix pMatrix and the view matrix (the view matrix depends on where the camera is and the direction in which it is looking). The composition of these matrices is named pvMatrix in the code.
If you want the opposite transform from the window back to three space, you have to take a clip space coordinate (x,y,z) and 'unproject' it back into 3D using the inverse of pvMatrix. In clip space, coordinates are in the range [-1,1], and the z coordinate is depth.
In the OpenGL world, these transforms are implemented in gluProject() and gluUnproject() (which you can Google for more information and source code).
In the jsfiddle example, we calculate the (x,y) coordinates in clip space, and then unproject (x,y,z) for two different values of z. From that we get two points in 3D space that map onto (x,y), and we can infer a direction vector. We can then move the camera in that direction to get a zoom effect.
In the code, the camera position is at the negation of the eye vector.
This example shows you how to move the camera in the direction that you are clicking. If you want to actually move towards specific objects in the scene, you have to implement something like object selection, which is a different kettle of fish. The example I've given is unaware of the objects in the scene.