Skydio camera and metadata overview

Each time you take a photo, you capture information that is referred to as metadata. This data is extremely useful when creating 3D reconstructions.

Metadata is stored in two types of files:

• EXIF (Exchangeable Image File): contains information about the specific camera settings used when capturing a specific image. This information includes details such as GPS position, camera make and model, date and time, shutter speed, focal length and more.
• XMP (Extensible Metadata Platform): the data standard for Adobe’s Extensible Metadata Platform

The information included in metadata files consists of camera intrinsics and camera extrinsics.

• Camera intrinsics refers to the focal length and optical center of the camera
• Camera extrinsics refers to the camera’s physical location and the direction it is pointing.

Using the extrinsic parameters, we can create camera coordinates which are mapped into the image plane using the intrinsics parameters.

In-depth

Skydio Camera Information

Photo Camera Intrinsics

Below are example values from EXIF/XMP data relevant to camera intrinsics. The values are from our “prototype” calibrations which should be very close for a drone of that type. Image files from a specific drone will have factory-calibrated values that will be even more accurate.

Skydio 2/2+ and X2 Color 

X2 Color/Thermal 

Computing pixel from point

The model we use is a simplification of the standard OpenCV model with the denominator coefficients and tangential coefficients omitted. This is a similar to the model from A Flexible New Technique for Camera Calibration, 1998. 

Below is an example of how to compute the projection of a point from the example Skydio 2 calibration in Python:

from __future__import print_function
import math

# Focal length = [Calibrated Focal Length X, Calibrated Focal Length Y]
F = [2376.5625, 2376.5626]
# Camera center = [Calibrated Optical Center X, Calibrated Optical Center Y]
C = [2027.5000, 1519.5000]
# Dewarp data
W = [0.13000, -0.24000, 0.10400]

# Example point in RDF camera frame.
# This point would be 1m right, 2m down, 5m forward of the camera. 
P = [1.0, 2.0, 5.0]

# Python function to compute pixel from point
def pixel_from_point(F, C, W, P): 
    """
    Example of gimbaled camera projection

    Args: 
       F: Focal lengths (pixels, length 2)
       C: Camera center (pixels, length 2)
       W: Distortion weights (length 3)
       P: RDF point in camera frame (length 3)

    Returns: pixel coordinates in form [column, row] 
    """
    P_img = [P[0] / P[2], P[1] / P[2]]
    r = math.sqrt(P_img[0] ** 2 + P_img[1] ** 2)
    weight = 1 + W[0] * (r ** 2) + W[1] * (r ** 4) + W[2] * (r ** 6)
    P_dist = [P_img[0] * weight, P_img[1] * weight]
    pixel = [P_dist[0] * F[0] + C[0], P_dist[1] * F[1] + C[1]]
    return pixel

pixel = pixel_from_point(F, C, W, P)
# Should print: [2511.003085, 2486.50617]
print(pixel)

Explore More

Explore related articles for additional information:

How to create 3D reconstructions

What Is Camera Calibration? by MathWorks

If you have questions or need additional support—feel free to reach out! Skydio Support

A0088