Case study: a circular antenna's 3D orientation from a single drone image

Summary: By treating the antenna as a planar circle, Vision Elements replaced two-view stereo matching with a single-image least-squares solve, giving the client a faster proof of concept and more accurate results.

The challenge

The client maps antenna poles by drone for maintenance and real-estate management; drone surveys cover large or hard-to-reach sites at lower cost and shorter turnaround. They needed each circular antenna's 3D orientation. Standard 3D reconstruction needs stereo: at least two shots from different positions and matching of the same points across both images, which is hard on an antenna edge. A homography could recover the pose from four control points, but that requires known 3D coordinates for each point, rarely available in the field. Vision Elements was engaged for a short-term project to find a way through.

What we did

Our computational scientists replaced point matching with geometric constraints. Assuming the antenna is planar and circular turns those properties into equations, so a single image is enough and no correspondence between views is needed. With the circular constraint, the eight homography parameters can be recovered from eight random point pairs rather than four known control points. Each image undergoes simple edge detection, and every edge point on the antenna rim contributes one equation. With N > 8 equations, a least-squares solve yields the antenna's orientation relative to the camera; more edge points strengthen the solution. Applying the camera's exterior orientation then expresses the antenna's orientation in the world coordinate system.

Outcome

The method produced more accurate results than the stereo-based state-of-the-art approaches the client had started from, and let the client build its proof of concept faster. The client's view was that the solution would not have been reached without this direction.

Why it matters for your project

Many inspection and survey problems look like they need more sensors, more images or more ground truth. Often they need a better formulation. If your pipeline is stalled on feature matching, missing control points or costly multi-view capture, a short engagement to re-pose the problem mathematically can remove the obstacle before you commit to hardware.

FAQ

Does it work with any camera? The solve gives orientation relative to the camera; converting to world coordinates uses the drone's camera-to-world exterior orientation.

What about antennas that are not circular? The method relies on the planar-circle assumption. A different shape would need its own constraint, which is the kind of reformulation we do.

How long was the engagement? It was a short-term engagement focused on the proof of concept.

Contact: info@vision-elements.com