transform

Use the transform model to apply transformations to input source images. The transformations are applied in the order they are written in the pipeline.

Configuration

{
  "name": "<your-camera-name>",
  "model": "transform",
  "api": "rdk:component:camera",
  "attributes" : {
    "source" : "<your-source-camera-name>",
    "pipeline": [
      { "type": "<transformation-type>", "attributes": { ... } },
    ],
    "intrinsic_parameters": {
      "width_px": <int>,
      "height_px": <int>,
      "fx": <float>,
      "fy": <float>,
      "ppx": <float>,
      "ppy": <float>
    },
    "distortion_parameters": {
      "rk1": <float>,
      "rk2": <float>,
      "rk3": <float>,
      "tp1": <float>,
      "tp2": <float>
    }
  }
}

Attributes

NameTypeRequired?Description
sourcestringRequiredname of the camera to transform.
pipelinearrayRequiredSpecify an array of transformation objects.
intrinsic_parametersobjectOptionalThe intrinsic parameters of the camera used to do 2D <-> 3D projections:
  • width_px: The expected width of the aligned image in pixels. Value must be >= 0.
  • height_px: The expected height of the aligned image in pixels. Value must be >= 0.
  • fx: The image center x point.
  • fy: The image center y point.
  • ppx: The image focal x.
  • ppy: The image focal y.
distortion_parametersobjectOptionalModified Brown-Conrady parameters used to correct for distortions caused by the shape of the camera lens:
  • rk1: The radial distortion x.
  • rk2: The radial distortion y.
  • rk3: The radial distortion z.
  • tp1: The tangential distortion x.
  • tp2: The tangential distortion y.

Pipeline transformations

The following transformation objects are available for the pipeline:

Classifications

Classifications overlay text from the GetClassifications method of the vision service onto the image.

{
  "source": "<your-source-camera-name>",
  "pipeline": [
    {
      "type": "classifications",
      "attributes": {
        "classifier_name": "<name>",
        "confidence_threshold": <float>,
        "max_classifications": <int>,
        "valid_labels": [ "<label>" ]
      }
    }
  ]
}

Attributes:

  • classifier_name: The name of the classifier in the vision service.
  • confidence_threshold: The threshold above which to display classifications.
  • max_classifications: Optional. The maximum number of classifications to display on the camera stream at any given time. Default: 1.
  • valid_labels: Optional. An array of labels that you to see detections for on the camera stream. If not specified, all labels from the classifier are used.

Crop

The Crop transform trims an image to a rectangular area specified by two points: the top left ((x_min, y_min)) and the bottom right ((x_max, y_max)). You can provide these points as integer pixel values or as decimal proportions of the image’s width and height. The origin ((0, 0)) occupies the top left pixel of the image; X values increase as you move right, Y values increase as you move down.

{
  "source": "<your-source-camera-name>",
  "pipeline": [
    {
      "type": "crop",
      "attributes": {
        "x_min_px": <int|float>,
        "y_min_px": <int|float>,
        "x_max_px": <int|float>,
        "y_max_px": <int|float>,
        "overlay_crop_box": <bool>
      }
    }
  ]
}

To crop a 100 x 200 image to the rectangular region between pixel coordinates (30, 40) and (60, 80), pass those coordinates in the following configuration:

{
  "source": "<your-source-camera-name>",
  "pipeline": [
    {
      "type": "crop",
      "attributes": {
        "x_min_px": 30,
        "y_min_px": 40,
        "x_max_px": 60,
        "y_max_px": 80,
        "overlay_crop_box": false
      }
    }
  ]
}

To crop any image to a rectangular region that occupies the central 50% of the image, use proportional coordinates (0.25, 0.25) and (0.75, 0.75):

{
  "source": "<your-source-camera-name>",
  "pipeline": [
    {
      "type": "crop",
      "attributes": {
        "x_min_px": 0.25,
        "y_min_px": 0.25,
        "x_max_px": 0.75,
        "y_max_px": 0.75,
        "overlay_crop_box": false
      }
    }
  ]
}

Attributes:

  • x_min_px: The X pixel or proportional value of the top left corner of the crop area.
  • y_min_px: The Y pixel or proportional value of the top left corner of the crop area.
  • x_max_px: The X pixel or proportional value of the bottom right point of the crop area.
  • y_max_px: The Y pixel or proportional value of the bottom right point of the crop area.
  • overlay_crop_box: When true, instead of cropping, overlays the cropping box on the original image to visualize where the crop would apply.

Detections

The Detections transform takes the input image and overlays the detections from a given detector configured within the vision service.

{
  "source": "<your-source-camera-name>",
  "pipeline": [
    {
      "type": "detections",
      "attributes": {
        "detector_name": string,
        "confidence_threshold": <float>,
        "valid_labels": ["<label>"]
      }
    }
  ]
}

Attributes:

  • detector_name: The name of the detector configured in the vision service.
  • confidence_threshold: Specify to only display detections above the specified threshold (decimal between 0 and 1).
  • valid_labels: Optional. An array of labels that you to see detections for on the camera stream. If not specified, all labels from the classifier are used.

Resize

The Resize transform resizes the image to the specified height and width.

{
  "source": "<your-source-camera-name>",
  "pipeline": [
    {
      "type": "resize",
      "attributes": {
        "width_px": <int>,
        "height_px": <int>
      }
    }
  ]
}

Attributes:

  • width_px: Specify the expected width for the aligned image. Value must be >= 0.
  • height_px: Specify the expected width for the aligned image. Value must be >= 0.

Rotate

The Rotate transformation rotates the image by the angle specified in angle_deg. Default: 180 degrees. This feature is useful for when the camera is installed upside down or sideways on your machine.

{
  "source": "<your-source-camera-name>",
  "pipeline": [
    {
      "type": "rotate",
      "attributes": {
        "angle_degs": <float>
      }
    }
  ]
}

Attributes:

  • angle_deg: Rotate the image by a specific angle in degrees.