CameraSnap enables the user to precisely calibrate camera mounting rotations (IMU->Camera rotations), receptor intrinsics, lens distortions, and individual image poses.

The amount and types of calibration required depends on the dataset. Aerial datasets may not require calibrating camera mounting rotations, but may benefit from individual image pose correction. Mobile datasets, on the other hand, almost always require mounting rotations to be calibrated, as the camera is typically removed from the vehicle (and thus removed from the IMU) after each collection.

Features and Matches

CameraSnap has 3 primary modes for handling feature matching, selectable from the topmost dropdown menu:

The options are:

  • Auto-detected without review: Use this for a fully-automatic calibration. This usually works well with aerial data sets.

  • Auto-detected with review: Use this to manually add, edit and remove feature matches when automatic calibration fails to produce good results. This is the best option for mobile/Ladybug5+ data sets.

  • Manually-created: This can be useful for troubleshooting and special datasets.

Choose the mode that makes sense for your application. For aerial data sets, you can probably calibrated the camera using Auto-detect without review. For mobile data sets, it's best to review matches before calibrating, to prevent erroneous matches from affecting the calibration.


Calibrating the camera does not require an entire data set worth of imagery. It's best to select a set of intervals specifically for calibrating the camera, to limit the number of images analyzed and ensure a proper calibration:

When making intervals for camera calibration, create intervals that:

  1. Include sections on the trajectory with opposing headings

  2. Include areas that contain man-made features, if possible

  3. Include areas where trajectory accuracy is high (good GNSS coverage)

  4. Avoid areas with high vehicle or pedestrian traffic

Regarding consideration #1: To properly calibrate the camera, you need images recorded facing in opposite directions. This could be imagery recorded along a U-turn, imagery recorded in opposing lanes of traffic, or ideally imagery recorded in a hashtag pattern:

Last updated