pixel_bearing

Contents

pixel_bearing#

rapidtools.processing.street_localization.pixel_bearing(x: float, y: float, camera_type: str = 'spherical', width: int | None = None, height: int | None = None, camera_parameters: Sequence[float] | None = None) → ndarray[source]#

Unit direction in the camera frame of a normalised image point.

Follows OpenSfM’s camera models, which is what Mapillary’s camera_type and camera_parameters describe. The camera frame has x to the right, y down and z forward.

Parameters:

  • x – Column as a fraction of the image width (0 left, 1 right).

  • y – Row as a fraction of the image height (0 top, 1 bottom).

  • camera_type – 'spherical' / 'equirectangular' for panoramas, 'perspective' / 'brown' for rectilinear images, or 'fisheye'. Unknown types are treated as perspective.

  • width – Image width in pixels. Needed with height for the aspect ratio; a 2:1 panorama or a 4:3 frame is assumed when unknown.

  • height – Image height in pixels.

  • camera_parameters – [focal, k1, k2] with the focal length normalised by the larger image side (OpenSfM convention). Ignored for spherical cameras; [0.85, 0, 0] is assumed when missing.

Returns:

numpy.ndarray – Unit vector (x, y, z) in the camera frame.

Example

>>> [round(float(v), 3) for v in pixel_bearing(0.5, 0.5)]        # pano centre
[0.0, 0.0, 1.0]
>>> [round(float(v), 3) for v in pixel_bearing(0.75, 0.5)]       # 90 deg right
[1.0, 0.0, 0.0]
>>> [round(float(v), 3) for v in pixel_bearing(0.5, 1.0)]        # straight down
[0.0, 1.0, 0.0]
>>> b = pixel_bearing(0.5, 0.5, 'perspective', 4000, 3000, [0.9, 0, 0])
>>> [round(float(v), 3) for v in b]
[0.0, 0.0, 1.0]
Contents