Camera

Camera

[Prerequisites]

Understood FUSEE's Scene Graph concept, especially how SceneNodeContainers and SceneComponentContainers work together.
Basic knowledge on matrix transformations in Computer Graphics (View and Projection matrices).

Using Cameras in the Scene Graph

Building a scene from code or loading a scene from a fus file will result in a SceneContainer, containing Nodes and Components.
To render the scene we can add a Camera to the SceneContainer. Note that this is no must but, for example, makes the implementation of follow or orbit cameras much easier (see the Examples section).

A camera consists of a SceneNodeContainer which contains a TransformComponent and a CameraComponent.

TransformComponent

As with every other TransformComponent the Translation, Rotation and Scale values are relative to the ones of the parent nodes. The cumulated Translation and Rotation values describe the cameras position in world space and therefore are the basis for calculating the View matrix (which is done in the background).

CameraComponent

The CameraComponent consists of the following properties:

Type	Name	Description
bool	Active	If set to true this camera generates a render output.
WritableTexture	RenderTexture	If this is not null the output gets rendered into the texture, otherwise to the screen.
float4	Viewport	The output is rendered to the section of the application window defined by the Viewport values (see table below).
float2	ClippingPlanes	Distance to the near (x value) and far (y value) clipping planes.
float	Fov	Vertical field of view in radians.
ProjectionMethod	ProjectionMethod	Sets the projection method. At the moment we can choose between perspective and orthographic.
int	Layer	If there is more than one CameraComponent in one scene, the rendered output of the camera with a higher layer will be drawn above the output of a camera with a lower layer.
float4	BackgroundColor	The background color of the render output.
bool	ClearColor	Is true per default. If set to false, the color bit won't be cleared before this camera is rendered. Set this to false if the background color should be transparent.
bool	ClearDepth	Is true per default. If set to false, the depth bit won't be cleared before this camera is rendered.
CustomCameraUpdate	CustomCameraUpdate	Is null by default. This delegate enables us to add a custom projection method. If we choose to use this delegate we need to provide a method that calculates a projection matrix and outputs a Viewport in percent in the range [0, 100]. Note that this is optional but if this delegate is not null its out values (Projection matrix and Viewport) will overwrite the ones calculated from the other camera parameters.

Viewport

The values are given in percent in the range [0, 100].

float4	x	y	z	w
	x value of the lower left corner	y value of the lower left corner	width	height

Examples

Creating the Camera SceneNodeContainer

The constructor of the CameraComponent expects a projection method, the distances to the clipping planes and the field of view in radians.
When creating the TransformComponent we can add a Scale value. This doesn't have an effect on the View matrix calculation, but can be useful if the camera node has a Mesh attached.

//1. Create a TransformComponent and a CameraComponent.
private TransformComponent _mainCamTransform;
private CameraComponent _mainCam = new CameraComponent(ProjectionMethod.PERSPECTIVE, 1, 1000, M.PiOver4);

public override async Task<bool> Init()
{
  _mainCam.Viewport = new float4(0, 0, 100, 100);
  _mainCam.BackgroundColor = new float4(1, 1, 1, 1);
  _mainCam.Layer = -1; 

  _mainCamTransform = _guiCamTransform = new TransformComponent()
  {
      Rotation = float3.Zero,
      Translation = new float3(0, 2, -10),
      Scale = new float3(0.33f, 0.33f, 0.5f)
  };

  [...]

  //2. Create the SceneNodeContainer.
  var cam = new SceneNodeContainer()
  {
      Name = "MainCam",
      Components = new List<SceneComponentContainer>()
      {
          _mainCamTransform,
          _mainCam,
          new Cube()
      },
      Children = new ChildList()
      {
          new SceneNodeContainer()
          {
              Components = new List<SceneComponentContainer>()
              {
                  new TransformComponent()
                  {
                      Scale = new float3(1, 1, 1),
                      Translation = new float3(0,0,-1.4f)
                  },
                  new Cube()
              }
          }
      }
  };

  //3. Add the camera to the scene.
  _rocketScene.Children.Add(cam);

  [...]

}

Orbit Camera

In the code below we use the method TransformComponent.RotateAround to create an orbit camera. This method expects a point (float3), that is used as the rotation center, and a rotation in angle-axis representation (float3 up-axis, float angle).

💡 Note: In the example the rotation center is (0, 0, 0), in practice it is useful to take the world space position of the object we want to rotate around. We can get this position for example by using SceneNodeContainer.GetGlobalTranslation().

public override void RenderAFrame()
{
    [...]

    //4 Use RotateAround(float3 center, float3 up-axis, float angle) to rotate the camera.
    var rotAxis = float3.UnitY * float4x4.CreateRotationYZ(new float2(M.PiOver4, M.PiOver4));
    _mainCamTransform.RotateAround(new float3(0,0,0), rotAxis, someAngle * DeltaTime);

    [...]  
}

💡 Note: We can also achive achieve an orbit camera by not rotating the camera itself, but by placing it under a SceneNodeContainer, which serves as a pivot point.
In this case we rotate the pivot around its origin. This method may prove more stable against cummulative rounding errors.

Follower Camera

A follower Camera can be achieved simply by adding the Camera SceneNodeContainer as a child to the object we want it to follow instead of putting it directly under the SceneContainer itself.

public override async Task<bool> Init()
{
  [...]

  //3. Add the camera to a node.
  _someParentNode.Children.Add(cam);
}

First Person Camera

To achieve a fps camera we can use TransformComponent.FpsView(). This method will modify the Translation and Rotation of the camera according to the parameters we pass.
It expects the following parameters:

Type	Name	Description
float	angleHorz	The horizontal rotation angle in radians. Should probably come from Mouse input (Mouse.XVel).
float	angleVert	The vertical rotation angle in radians. Should probably come from Mouse input (Mouse.YVel).
float	inputWSAxis	The value we want to translate the camera when pressing the W or S key (Keyboard.WSAxis).
float	inputADAxis	The value we want to translate the camera when pressing the A or D key (Keyboard.ADAxis).
float	speed	Changes the speed of the camera movement.

public override void RenderAFrame()
{
    [...] 

    _camTransform.FpsView(_angleHorz, _angleVert, Keyboard.WSAxis, Keyboard.ADAxis, Time.DeltaTime * 1000);

    [...]
}

View and Projection matrices without a SceneGraph

If we render without a scene graph or do not want to use a camera for whatever reason, we can set the Viewport, Projection matrix and View matrix directly on the RenderContext.

public override void RenderAFrame()
{
    RC.Viewport(0, 0, Width, Height);
    [...] 

    _camTransform.FpsView(_angleHorz, _angleVert, Keyboard.WSAxis, Keyboard.ADAxis, Time.DeltaTime * 1000);

    [...]

    var mtxRot = float4x4.CreateRotationX(_angleVert) * float4x4.CreateRotationY(_angleHorz);
    var mtxCam = float4x4.LookAt(0, 20, -_zoom, 0, 0, 0, 0, 1, 0);

    var view = mtxCam * mtxRot;
    var perspective = float4x4.CreatePerspectiveFieldOfView(_fovY, (float)Width / Height, ZNear, ZFar) * mtxOffset;

    RC.View = view;
    RC.Projection = orthographic;

    [...]
}

If we don't set the parameters at all FUSEE will use predefined default values. Those can be read from RenderContext.DefaultState.

Implementation details

👷 Engine Developer

In order for the camera to work in the expected way, each SceneRenderer – whether it is a Forward, Deferred or a custom one – needs a pre-rendering pass (one traversal of the scene). This pre-pass is responsible for collecting the CameraComponents and calculating the view matrix for each one.

This is implemented in the PrePassVisitor. For each CameraComponent that is found while traversing the scene, we calculate the View matrix from the cumulated Model matrix by eliminating its scale (get the scale factor and divide the 3x3 rotation/scale part of the matrix by it) and inverting it (view = _state.Model.Invert()).
The CameraComponent and its View matrix is then saved as a CameraResult to a public list and can therefore be accessed from the SceneRenderer.

Back in the scene renderer we render one time for each camera we've found in the pre-pass, taking the CameraComponents' Viewport the resulting Projection matrix and the CameraResults' View matrix into account.