3d engine redesign

[ferdymania]

hi guys
i have decided to refactor and start again doing it properly..
the process will be..
--------------------------
1. requirements doc
2. use cases
3. drawing pictures and lines (concepts)
4. class diagrams
5. crc cards..
6. interaction diagrams
7 algorithms
8, test first coding
9. Re iterate..
-----------------------------

here is the very first rough as guts req doc.

Requirements for 3d engine
----------------------------------------
import .obj models
rotation x,y,z
translation x,y,z
terrain generation
camera
omnidirectional lighting + ambient lighting
flat shading, gouraud shading, texture mapping
collision detection
frustum culling
backface culling
Projection to 2d screen.
render to screen

-------------------------------------------------

please add comments and thoughts ...
i want to avoid feature creep at this stage..
although anything vital please kik me.. :-)

thank you
steve
<- guy sebastian , my mentor

[BlinkOk]

i've found this usefull from time to time;
http://home.planet.nl/~monstrous/

[ferdymania]

thanks blink .. i went to that site and it is awesome..

now to my revised doc.. please tell me if i am being unrealistic..

Requirements for 3d engine
----------------------------------------
+ IMPORT .obj MODELS
allow extensibility to add normals, materials, animation etc
allow capacity to extend to formats such as asc, md2, 3ds
+ DATA STRUCTURES must be fast
Linked lists - see watt appendix
Octrees
BSP partitioning
+ ROTATION x,y,z
+ TRANSLATION x,y,z
+ TERRAIN generation
Sinusoidal
Fractal
Other ? heightmap
+ CAMERA
+ Omnidirectional LIGHTING + ambient lighting, distance- inverse %
+ flat SHADING, gouraud shading, texture mapping
+ BOUNDING BOXES - AABB ?
+ COLLISION detection
How to detect collision with terrain?
+ FRUSTUM culling
+ BACKFACE culling
+ OPTIMISE code heavily
Precalculate
Single loop rendering and calcs..
+ PROJECTION to 2D screen.
+ RENDER/DRAW to screen

-------------------------------------------------

[BlinkOk]

probably unrealistic for flash! however i been thinking you could make it a 50/50 deal. www.screenweaver.com is a screen saver type thingy. it allows you to call .dll's from flash. you might be able to code all the 3d stuff in C and just use flash as the render. it would all depend on the speed that screenweaver call's .dll's but i think it would be worth a try.
btw. this is another pretty good site;http://www.flipcode.com/

[ferdymania]

hi blink.. i thought i was a bit unrealistic..
i will have to cut it down a bit..
although.. if i precalculate a lot.. hmm..
probably ditch other shading and stick with flat..
and have a very simple lighting equation..
i think the main restriction is the number of faces..
probably the max might be about 300 faces..
so it will be clunky...
but anyway.. i want the design to be non platform specific
and i might code one in java as well...
anfy3D is pretty awesome..
and i been to flipcode its good..

go the fro'

[ferdymania]

and here are my
----------
use cases
-----------
+ Import Model Data
- Assign the Vertices to a pointsList Data Structure ,
- assign Faces to a faceList, normals?, colours?
- use external pHP script
+ Create Models
- From the data, create the model object and
- assign the data to object class variables.
+ Rotate Model x y z
+ Translate Model x y z
+ Import Terrain Data
- Have the ability to parse terrain data from external sources
- use Php script
+ Create Terrain
Assign data to terrain class object variables
+ Place Terrain in World
- set position and relation to the engine world system
+ Place Models in World
- Position static and moveable objects in initial positions
+ Define Moveable and static Object Types
- Moving Objects - planes, tanks, cars, spaceships
- Static Objects - Terrain, Buildings , trees, Fortress that fires at you(eg)
+ Create Camera
+ Set Camera to follow Object
+ Change Camera settings
- POV, follow another object
+ Light any/all Objects
+ Shade objects
+ Create Bounding Boxes - Detect Collisions
+ Perform Backface Culling
+ Occlusion Culling ??
+ Convert to 2D Coordinates
+ Draw/Render to screen
+ Z-Sort Objects
- Flash - use swapDepths
- Get average Z for face - Z sort routine

[BlinkOk]

problem is, any loop for 300 odd iterations, even if it is doing nothing will slow the flash player. i think the primary objective is to somehow reduce the size of the table.
i always thought (in a landscape sitution) you could reduce the table by making it kinna tile based in that the player's depth of vision is limited to those polys that are directly in front of him.

[ferdymania]

me and milo have been testing linked lists and he has just
written a method for using arrays by pushing and popping..
the linked lists are 50% faster than conventional arrays..
and milo's arrays are about the same.. we have done tests on several hundred points and even thousands... several hundred is realistic..
thousands of course not... doh!
i did a 700 face poly and it was a slug.. so that seems the extreme upper limit.. but i did that before the list was optimised...
i have got some more tests to do tomorrow..
and mx2004 is even faster i am sure..

probably your idea is good about excluding vision.. i guess cull
anything beyond say a couple of thousand z.. and use inverse proportional distance lighting.. to hide it..
use frustum culling that is extreme maybe.. ???
how did you do your 3d demo .. the columns in that room?
i saw strilles raycasting demo and that was pretty good..
the doom-like one..
Where did you guys get to with your 3d stuff.?. did you think it was not feasible.. ??
anyway... sleep beckons
steve

[ferdymania]

Here is the classes and responsibilities

Vector3D
--------
Store Coordinates
Able to do Vector maths , + - * /

Matrix
------
Hold Matrix data
create identity matrix
Create rotation/translation matrices
Create Scale/Shear/Deformation matrices
Allow for matrix multiplication/concatenation

Face
-----
Store list of points
store the normal, colour
Set visibility
Do occlusion, frustum and backface culling

Mesh
----
Store list of faces
set centrepoint
set radius of bounding sphere
Set moving flag

Terrain
-------
Set up grid
set heightmap
set texture or colour
find nearest neighbour - interpolate/extrapolate heights
draw it.

Light
------
Hold the vector
set brightness
set ambience
find distance from models

Camera
-------
Set it up
Attach it to a model
set position
reset model to look at
set a radius or viewAngle

Engine
------
Oversee the engine
interface or plugin for developers
create models
set to render
Allow input for players
Initialise Terrain and unmoving Models
-------------------------------------------

Next up is working all the class definitions,
set the member variables and methods , what arguments they will need
and their return types..


thanks guy

[03ricr]

i've been working on terrain.. from scratch with only some scans of papers on ROAM and LOD etc.. so far it looks good but if you want you're engine to be faster and better quality than a normal sorta terrain grid u should look at my ideas ferdy... hopefully i will work more on them after i get this light stuff worked out.

[ferdymania]

hi ross
your terrain sounds great and i will certainly use your ideas..
i havent even thought about how to implement LOD...
the lighting calcs can be used in the same loop to calculate
backface culling and frustum culling using the face normal..
and the light vector...
i still havent had a look at your code yet..
i been so busy ... sorry mate..
hopefully this weekend..
steve

[03ricr]

the good thing about the terrain is you can pre-calculate the light becuase the terrain and the light always stay the same.

[ferdymania]

aha... good thingking 99
of course..
back to the grindstone...