It''s not that difficult as you think.
This is not a new invention, even Monkey island use this kind of gameplay.

Well, i don''t EXCACTLY know how they do it, but this how i would do:


Say you have a scene on a street. you got a road going from left to right, you got an ally going straight "into" the image, and a bit to the left there''s a car parked.

We want to control a person that can walk on the street and into the ally, but not into the building (That would look weird), nor the car.

First, we need some kind of koordinate system to base everything on. we take the X axis going from left to right, and Y going "into" the image.
Now you have a person that walks in the scene the right way.
If he goes "deeper" into the image he gets smaller, and so on.

Now, the only thing missing now is the collision detection.
We have the koordinates for the person. Now we create a "virtual" version of the scene, which is described top down, where we tell where the person can and can''t go.

For example:

OOOOOO^OOOOOOO
OOOOOO OOOOOOO
OOOOOO OOOOOOO
< ssss x >
< OOO >
< >
OOOOOOOOOOOOOO

where O''s are areas where the person can''t go, the "arrows" are triggers which says "change scene" or something, and x is the position of the person. The person is stanging on nothing, so he can stand there, but if goes a step up, the "virtual" scene says no.

That''s about it. Of course, there''s a lot of issues to take care of. for example. the person can''t walk on the car, but he can walk BEHIND the car. This means that the program must "feel" when the person is present behind the car (on the s''es,for example), and draw the car separately at that point.

Electron

"Your truth can be changed simply by the way you accept it."

if i were to use this method of rendering, i''d probably use
a vector collision network..
its pretty simple in theory, i''ve used it for irregularly shaped
collision networks in direct draw, i''d assume it''d be
equally simple in any other API..
for this type of collision network, you''d essentially create a
tool that would display your image, or "map", and you''d just
click away drawing lines around what would be solid or unpassable
objects.
to test for collision, in your engine you''d simply test if the lines
of the vector network and the lines surrounding your character''s
impassable bits have intersected or not.
there are probably easier ways to do it (like hidden polygons, ect),
this is just the method i prefer since i''ve done it a few times.

-eldee
;another space monkey;