glTF Perspective Camera Projection Parameters

In glTF (GL Transmission Format), a camera node defines a perspective projection by referencing a reusable perspective camera object stored in the asset's root arrays. This article explains how glTF separates spatial transformation from camera optics, details the specific parameters within the perspective property—including vertical field of view, clipping planes, and aspect ratio—and shows how these values construct standard projection matrices.

Separation of Nodes and Camera Definitions

glTF separates the camera's spatial position from its optical characteristics:

  • The Node: Defines the camera's transform (translation, rotation, and scale) in the 3D scene graph. A node points to a camera by its index in the root cameras array using the "camera" property.
  • The Camera Object: Defines the projection math. It resides in the top-level "cameras" array and specifies a "type" (set to "perspective") and a "perspective" object containing the projection parameters.

By default, a glTF camera looks down its local negative Z-axis (-Z), with the local Y-axis pointing up (+Y) and the local X-axis pointing right (+X).

Perspective Camera Properties

Inside the "perspective" object, glTF defines four core attributes:

{
  "cameras": [
    {
      "type": "perspective",
      "perspective": {
        "aspectRatio": 1.5,
        "yfov": 0.785398,
        "zfar": 100.0,
        "znear": 0.1
      }
    }
  ],
  "nodes": [
    {
      "camera": 0,
      "translation": [0.0, 1.0, 5.0]
    }
  ]
}

  1. yfov (Float, Required): The vertical Field of View specified in radians. For example, a 45-degree vertical FOV is represented as roughly 0.785398 radians. glTF uses vertical FOV rather than horizontal FOV as the primary metric, matching common graphics API conventions.

  2. znear (Float, Required): The distance from the camera origin to the near clipping plane along the negative Z-axis. This value must strictly be greater than zero (znear > 0) to prevent division-by-zero errors in perspective divide operations.

  3. zfar (Float, Optional): The distance from the camera origin to the far clipping plane along the negative Z-axis. When present, it must be greater than znear. If omitted, the camera defines an infinite perspective projection, where the far plane extends indefinitely.

  4. aspectRatio (Float, Optional): The aspect ratio of the viewport, defined as width divided by height (width / height). This property is optional; if omitted, the rendering engine dynamically computes the aspect ratio from the display canvas dimensions.

Matrix Derivation

When a runtime engine parses these parameters, it computes a standard right-handed perspective projection matrix:

  • Horizontal Field of View: Computed as xfov = 2 * atan(tan(yfov / 2) * aspectRatio).
  • Finite Projection: When zfar is present, the matrix maps znear and zfar to standard normalized device coordinates (typically [-1, 1] or [0, 1] depending on the graphics backend like WebGL, Vulkan, or DirectX).
  • Infinite Projection: When zfar is undefined, the engine uses the infinite perspective projection formulation, avoiding depth-buffer precision issues at extreme distances.