Skip to content

Latest commit

 

History

History
87 lines (76 loc) · 5.17 KB

README.md

File metadata and controls

87 lines (76 loc) · 5.17 KB
Raw

Camera Capture Plugin for UE 5

Plugin for capturing camera (GB + depth + motion vectors) data from multiple cameras in UE 5

Provides:

  • An actor component CaptureComponent which can be added to any actor. It will automatically find all SceneCapture2D components that are attached to the parent actor. For each camera it finds, it will configure that camera to output depth and motion vectors (by applying the M_DmvCapture post process material to it) and will configure it to output to a render texture. It will then create a copy of each camera which will render the color data to a texture as well. These textures can be captured (for other uses in the engine), and optionally serialized / saved to disk. The component can be configured to capture every frame of the camera or to capture frames at a specific interval of time.
  • A M_DmvCapture material for saving depth + motion vectors to a single texture (used by CaptureComponent)

See the unreal-camera-capture-example repo for a simple example project using this plugin.

See also the unreal-python-tools repo for some example code which provides processing and display of the data produced by the component in this repo.

Data Collected

The CameraCapture plugin's CaptureComponent creates the following data files within the SaveLocation:

  • camera_config.csv: CSV formatted data (first row is header), which as the follow information:
    • name: The name of the camera
    • width: The width of the image in pixels
    • height: The height of the image in pixels
    • focalLength: The focal length of the camera in meters
    • fov: The horizontal field of view of the camera
    • nearClipPlane: The near clip plane of the camera view frustum in meters
    • farClipPlane: the far clip plane of the camera view frustum in meters (may be inf)
    • tx: The relative position of the camera w.r.t. the actor position, x-axis
    • ty: The relative position of the camera w.r.t. the actor position, y-axis
    • tz: The relative position of the camera w.r.t. the actor position, z-axis
    • qw: The relative orientation of the camera w.r.t. the actor orientation, w-component
    • qx: The relative orientation of the camera w.r.t. the actor orientation, x-component
    • qy: The relative orientation of the camera w.r.t. the actor orientation, y-component
    • qz: The relative orientation of the camera w.r.t. the actor orientation, z-component
  • transformations.csv: CSV formatted data (first row is header), which has the following information:
    • i: frame index
    • time: time since the start of the program at which data (frames/transforms) were captured
    • tx: actor world position along the world x-axis
    • ty: actor world position along the world y-axis
    • tz: actor world position along the world z-axis
    • qw: actor world orientation as a quaternion, w-component
    • qx: actor world orientation as a quaternion, x-component
    • qy: actor world orientation as a quaternion, y-component
    • qz: actor world orientation as a quaternion, z-component
  • <camera name>_<index>.raw: Uncompressed 4-channel, 32bit float formatted (32bit float per channel) color image data for <camera_name> at 0-indexed <index>. The mapping from <index> to time can be found in the transformations.csv file.
  • <camera name>_depth_motion_<index>.raw: Uncompressed 4-channel 32bit float formatted (32bit float per channel) depth (red channel) + motion vector data (pixel x-y motion frame to frame in the green and blue channels) for <camera name> at 0-indexed <index>. The mapping from <index> to time can be found in the transformations.csv file.

The output data in this folder can be visualized using the display_raw python script. An example visualization can be found below:

example

References