flat assembler
Message board for the users of flat assembler.
Index
> Heap > camera to view matrix 
Author 

tthsqe
I like to keep the camera orientation as a unit quaternion.
Recall that the unit quaternions form a double cover of SO(3). (q and q map to the same matrix) The problem with maintaining the camera orientation directly as an orthogonal matrix M is that round off errors will occur when you multiply M by other rotations, and then it is non trivial to get M back to as close to orthogonal as possible. However, it is trivial to renormalize a quternion, and matrix multiplication corresponds to quaternion multiplication. I think it is elegant, and you can see how I do it in 'camera.inc' of my post in http://board.flatassembler.net/topic.php?t=15746&start=0 

18 Feb 2014, 21:32 

gens
i didn't learn quarterions at school (nor matrices, as far as i remember)
so that will take me time the camera is for now as an object, absolute world placement Code: camera_position: dd 0f, 0f, 0f camera_rotation: dd 0f, 0f, 0f (for now) where rotation is rotations around x,y,z in radians then before rendering a frame i plan to calculate a matrix out of it and multiply it by object's matrices (that are also calculated; now for every frame, later will be only when changed) thx for the example i see there is plenty theory there to learn bdw i remember you wanting to render many triangles now i mostly understand opengl buffers and rendering calls terminology well enough to make a quick example one of my test meshes has 28208 triangles and its np for cpu/gpu (glDrawArrays) i could also give the whole thing, but it's still work in progress (ugly at places) 

18 Feb 2014, 22:40 

revolution
gens wrote: where rotation is rotations around x,y,z in radians 

18 Feb 2014, 22:45 

gens
revolution wrote:
ye i imagined making it like that when starting, don't remember why it went this way but i need tilting anyway so its good i got 3 separate functions for making a rotational matrix around x,y and z will make a different function that combines them, but they will still be useful for like two to two and a half dimensional games 

18 Feb 2014, 22:53 

tthsqe
I have several objection to characterizing your camera orientation by three angles.
First, suppose you want to rotate your camera by an angle theta about the vector {x,y,z}. What are the new three angles? Nobody wants to work out those calculations, and they are literally 'transcendental' in difficulty. Second, as the angles really only matter mod Pi or mod 2*Pi, the correspondence between angles and camera orientations is manytoone. With unit quaternions, the correspondence is only twotoone. Also, the rotation problem can be solved my multiplying on the left by Code: cos(theta)+sin(theta)*(x*I+y*J+z*K)/sqrt(x^2+y^2+z^2) Very simple. 

19 Feb 2014, 04:29 

gens
tthsqe wrote: First, suppose you want to rotate your camera by an angle theta about the vector {x,y,z}. What are the new three angles? Nobody wants to work out those calculations, and they are literally 'transcendental' in difficulty. sounds fun tthsqe wrote: Second, as the angles really only matter mod Pi or mod 2*Pi, the correspondence between angles and camera orientations is manytoone. rotation is sanitized when changed (if over 2pi, sub 2pi) figured i don't want overflows, no matter how far fetched tthsqe wrote: With unit quaternions, the correspondence is only twotoone. Also, the rotation problem can be solved my multiplying on the left by its back to the quarterion FAQ for me thx for the tips il' implement quarterions when i understand them for now i'l limit the camera to two rotations so i can work on more fun parts (plan is a generic engine, so multiple types of objects, and now cameras, will be ok) 

19 Feb 2014, 15:19 

cl0ckw0rkz14
I didn't get taught matrices at school either and have never really been confident at mathematics but this tutorial helped me loads when I was writing a 2d game for android. http://www.opengltutorial.org/beginnerstutorials/tutorial3matrices/


19 Feb 2014, 22:29 

gens
yes it's a good tutorial but it is cpp with external libraries
http://www.cs.princeton.edu/~gewang/projects/darth/stuff/quat_faq.html this one has lots of good maths stuff for 3D 

20 Feb 2014, 23:04 

tthsqe
gens, is my example 'camera.inc' unreadable? These links seem to make a dog's dinner of the issues. My 'camera.inc' everything one could possibly want for a camera with free motion in all of its six dimensions.
Data structures: CameraPos: {x,y,z} CameraQuat: {x,y,z,w} with x^2+y^2+z^2+w^2=1 Given these two data structures, I have code to  convert CameraPos and CameraQuat to a matrix for opengl modelview matrix  rotate CameraQuat about a given vecotor by a given angle (3 dimensions for camera rotation)  move CameraPos relative to a given vector (3 more dimensions) I allow each of these six degrees of freedom 2 keys for a total of 12 key to control the camera. You can avoid the lameo c++ examples and all of their bloat. 

20 Feb 2014, 23:42 

gens
the example is good, i will surely use it as a reference
thing is i don't understand yet why quaterions work so i don't know their limitations/advantages like when i come to physics where there will be lots of rotations and changes of same, i have to know what is going on now i don't even know how to point at something using quaterions all functions are good to have like many fps don't do roll on characters (maybe some on the camera but not things like bounding boxes, from what i know) strategies... they many times don't even have a third dimension internally (sc2 has like fixed height plateaus (weird word that)) and ofc all code no libraries other then 3 that i actually need (for window, for loading images and for GL ofc) savagexr is one of the inspirations when planing in that game teams can be 30v30 players and it still runs fine simple physics, but you can swing around you with claws/swords and every hit is well timed over the network it's a great game 

21 Feb 2014, 04:47 

tthsqe
Quote: thing is i don't understand yet why quaterions work so i don't know their limitations/advantages A quaternion is a number of the form w + x I + y J + z K where {w,x,y,z} are real. Multiplication of a quaternion by real numbers is commutative. They are multiplied as in any ring with the equivalences I*I = J*J = K*K = I*J*K = 1 With these rules, you can write down the four components of any product of quaternions. Now, the useful thing is that to each unit quaternion (w + x I + y J + z K with w^2+x^2+y^2+z^2=1) we can associate the rotation matrix Code: ; w^2+x^2y^2z^2 2 (x yw z) 2 (w y+x z) ; 2 (x y+w z) w^2x^2+y^2z^2 2 (y zw x) ; 2 (x zw y) 2 (w x+y z) w^2x^2y^2+z^2 As I mentioned before this gives a 2to1 covering of the rotation matrices. And matrix multiplication is compatible with quaternion multiplication, as you can check by tedious calculation. To solve your 'look at' problem, lets assume that we are at a point {x,y,z} that is a distance of 1 away from the origin. What is quaternion associated with the matrix that will point us to the origin? It is simply Code: (x*I + y*J + (z + 1)*K)/Sqrt[2 z + 2] (you can also multiply this on the right by any quaternion of the form cos(t)+sin(t)*K to roll your camera) You can verify this by computing the associated rotation matrix and finding out that the last column is {x,y,z} (assuming x^2+y^2+z^2=1), and this is exactly what my camera.inc says it needs to be. 

21 Feb 2014, 08:03 

gens
so i thought about it and slept on it
quaternion (i always get the name wrong) rotations is rotating around a normalized vector x,y,z by w radians ? 

22 Feb 2014, 13:28 

tthsqe
gens,
almost. The rotation matrix associated with the unit quaternion {w,x,y,z} is a rotation about the vector {x,y,z}, but the angle is 2*arccos(w). 

22 Feb 2014, 18:23 

gens
kind of makes sense now, il have to put some things on paper first
i'm more of a mechanics person then mathematics from what i read the inventor of quaternions was also, so il look into hes writings later thx 

23 Feb 2014, 00:56 

< Last Thread  Next Thread > 
Forum Rules:

Copyright © 19992020, Tomasz Grysztar.
Powered by rwasa.