---
title: "Comment by Ahmet Özel on \"A Camera Is Not a GPS Receiver: How Visual-Inertial Odometry Helps Drones Navigate\" | Hashnode"
description: "The coordinate-frame check should include uncertainty as well as position and orientation. Rotating a pose into the aircraft frame while leaving an anisotropic"
url: "https://hashnode.com/posts/visual-inertial-odometry-drone-navigation/6aed707b1b4fdc91da515196/comment/6af1576f3097fe7d1baf9b23"
---

[Discussion](https://hashnode.com/posts/visual-inertial-odometry-drone-navigation/6aed707b1b4fdc91da515196)

Comment [A Camera Is Not a GPS Receiver: How Visual-Inertial Odometry Helps Drones Navigate](https://hashnode.com/posts/visual-inertial-odometry-drone-navigation/6aed707b1b4fdc91da515196)

[Ahmet Özel](https://hashnode.com/@ahmetozel)

[2d ago](https://hashnode.com/posts/visual-inertial-odometry-drone-navigation/6aed707b1b4fdc91da515196/comment/6af1576f3097fe7d1baf9b23)

The coordinate-frame check should include uncertainty as well as position and orientation. Rotating a pose into the aircraft frame while leaving an anisotropic covariance in the camera frame changes which direction the estimator believes is reliable, even when the pose message looks plausible.

A replay fixture with a known mounting rotation and much greater uncertainty along one camera axis would make this visible. The transformed pose and uncertainty should agree about the same physical direction. That would test a part of the integration contract that a straight-line trajectory with symmetric uncertainty can accidentally hide.