In image-based 3D modelling, we assume that all light rays forming an image pass through a single perspective centre at a single instant.

But most CMOS cameras violate that assumption because they expose the sensor row by row (electronic rolling shutter), meaning that each sensor line is captured at a slightly different time. The speed of this process, which I term the shutter transit time, typically ranges from 1/10 s to 1/60 s in consumer cameras.

This paper shows that such transit times can significantly degrade camera orientation and 3D mesh quality, even when standing still and shooting handheld using very short exposure times.

Software-based rolling-shutter compensation helps, but the result never reaches the geometric fidelity of a fast mechanical rolling-blind shutter.

To support deeper analysis of this issue in the future, the paper also introduces a robust shutter classification framework and standardised terminology for the key temporal quantities in the image-formation pipeline.

https://isprs-archives.copernicus.org/articles/XLVIII-2-W12-2026/495/2026/