Skip to content

18/03/2026 10:28 AM - PACES

⬅️ [17/03/2026 10:41 AM - PS3](<./17_03_2026 10_41 AM - PS3.md>) | ⬆️ [EECS 504](<./README.md>) | [18/03/2026 4:42 PM - Lecture](<./18_03_2026 4_42 PM - Lecture.md>) ➡️

18/03/2026 10:28 AM - PACES
Aidan Dempster - adempst - 6125 9596

Problem

Given a single perspective image and a minimal set of known measurements in the scene, how can we recover information about a broad set of other measurements in the scene?

Approach

The paper proposes a geometric framing of a set of questions that can be answered based on knowledge of a reference plane and vanishing point that does not lie parallel to the reference frame. This then allows for measurements on the reference plane, measurements on planes parallel to the reference plane (as long as we have an additional measurement), and measurement of camera position. The paper also provides methods to analyze the confidence in predictions.

Contribution

The paper provides presumably novel formulas to recover measurements from single view perspective images using a reference plane and a vanishing point. They provide derivations of the formulations and derivations for confidence measures.

Evaluation

A single image is provided that uses their method to measure the height of an individual, which is close to the true height. However, the true evaluation is the the derivations which prove that given the assumptions are met, the measurements can be made exactly.

Substantiation

The proofs provided are very convincing that this method is valid. However, I would appreciate more sensitivity analysis with respect to camera parameters. The uncertainty analysis provided is with respect to measurements in the scene which is important, but calibration generally is not perfect distortion parameters would effect the results.

single_view_metrology.pdf


⬅️ [17/03/2026 10:41 AM - PS3](<./17_03_2026 10_41 AM - PS3.md>) | ⬆️ [EECS 504](<./README.md>) | [18/03/2026 4:42 PM - Lecture](<./18_03_2026 4_42 PM - Lecture.md>) ➡️