Skybox旋转与世界不同步

Question

I've got a small scene with a loaded mesh, a ground plane and a skybox. I am generating a cube, and using the vertex positions as the cubemap texture co-ordinates.

Horizontal rotation (about the y-axis) works perfectly and the world movement is aligned with the skybox. Vertical rotation (about the camera's x-axis) doesn't seem to match up with the movement of the other objects, except that the strange thing is that when the camera is looking at the center of a cube face, everything seems aligned. In other words, the movement is non-linear and I'll try my best to illustrate the effect with some images:

First, the horizontal movement which as far as I can tell is correct:

Facing forward:

Facing left at almost 45Deg:

Facing left at 90Deg:

And now the vertical movement which seems to have some discrepancy in movement:

Facing forward again:

Notice the position of the ground plane in relation to the skybox in this image. I rotated slightly left to make it more apparent that the Sun is being obscured when it shouldn't.

Facing slightly down:

Finally, a view straight up to show the view is correctly centered on the (skybox) cube face.

Facing straight up:

Here's my drawing code, with the ground plane and mesh drawing omitted for brevity. (Note that the cube in the center is a loaded mesh, and isn't generated by the same function for the skybox).

void MeshWidget::draw() {

    glMatrixMode(GL_MODELVIEW);
    glLoadIdentity();

    glPushMatrix();

    glRotatef(-rot_[MOVE_CAMERA][1], 0.0f, 1.0f, 0.0f);
    glRotatef(-rot_[MOVE_CAMERA][0], 1.0f, 0.0f, 0.0f);
    glRotatef(-rot_[MOVE_CAMERA][2], 0.0f, 0.0f, 1.0f);

    glDisable(GL_DEPTH_TEST);
    glUseProgramObjectARB(shader_prog_ids_[2]);
    glBindBuffer(GL_ARRAY_BUFFER, SkyBoxVBOID);
    glEnableClientState(GL_VERTEX_ARRAY);
    glVertexPointer(3, GL_FLOAT, sizeof(Vec3), BUFFER_OFFSET(0));

    glBindBuffer(GL_ELEMENT_ARRAY_BUFFER, SkyIndexVBOID);
    glDrawElements(GL_TRIANGLES, 36, GL_UNSIGNED_SHORT, BUFFER_OFFSET(0));
    glUseProgramObjectARB(shader_prog_ids_[0]);
    glEnable(GL_DEPTH_TEST);

    glPopMatrix();

    glTranslatef(0.0f, 0.0f, -4.0f + zoom_factor_);

    glRotatef(rot_[MOVE_CAMERA][0], 1.0f, 0.0f, 0.0f);
    glRotatef(rot_[MOVE_CAMERA][1], 0.0f, 1.0f, 0.0f);
    glRotatef(rot_[MOVE_CAMERA][2], 0.0f, 0.0f, 1.0f);

    glPushMatrix();

    // Transform light to be relative to world, not camera.
    glRotatef(rot_[MOVE_LIGHT][1], 0.0f, 1.0f, 0.0f);
    glRotatef(rot_[MOVE_LIGHT][0], 1.0f, 0.0f, 0.0f);
    glRotatef(rot_[MOVE_LIGHT][2], 0.0f, 0.0f, 1.0f);

    float lightpos[] = {10.0f, 0.0f, 0.0f, 1.0f};
    glLightfv(GL_LIGHT0, GL_POSITION, lightpos);

    glPopMatrix();

    if (show_ground_) {

        // Draw ground...
    }

    glPushMatrix();

    // Transform and draw mesh...

    glPopMatrix();
}

And finally, here's the GLSL code for the skybox, which generates the texture co-ordinates:

Vertex shader:

void main()
{
    vec4 vVertex = vec4(gl_ModelViewMatrix * gl_Vertex);

    gl_TexCoord[0].xyz = normalize(vVertex).xyz;
    gl_TexCoord[0].w = 1.0;

    gl_TexCoord[0].x = -gl_TexCoord[0].x;

    gl_Position = gl_Vertex;
}

Fragment shader:

uniform samplerCube cubeMap;

void main()
{
    gl_FragColor = texture(cubeMap, gl_TexCoord[0]);
}

I'd also like to know if using quaternions for all camera and object rotations would help.

If you need any more information (or images), please ask!

Answer 1

I think you should be generating your skybox texture lookup based on a worldspace vector (gl_Vertex?) not a view space vector (vVertex).

I'm assuming your skybox coordinates are already defined in worldspace as I don't see a model matrix transform before drawing it (only the camera rotations). In that case, you should be sampling the skybox texture based on the worldspace position of a vertex, it doesn't need to be transformed by the camera. You're already translating the vertex by the camera, you shouldn't need to translate the lookup vector as well.

Try replacing normalize(vVertex) with normalize(gl_Vertex) and see if that improves things.

Also I might get rid of the x = -x thing, I suspect that was put in to compensate for the fact that the texture was rotating in the wrong direction originally?

I'd also like to know if using quaternions for all camera and object rotations would help.

Help how? It doesn't offer any new functionality over using matrices. I've heard arguments both ways as to whether matrixes or quaternions have better performance, but I see no need to use them.