Rectilinear perspective is a model of reality. It is useful, but like all models [it is wrong][1]. The key is that you have to "look around" as you draw.

Human foveal vision is restricted to the an angle of about 5 degrees and the that five degrees gets sharper toward the center so the zone of high sharpness is even narrower. The human eye also contains a blind spot where the optical nerve connects in lieu of rods and cones. This means that our eyes scan around a scene and our brains form sharp cohesive mental images. [a] The sample image works when our eye scans around it looking at details. It crumbles when viewed as a gestalt.

The utility of a rectilinear perspective rendering is that it models [b] part of that cohesive mental image reasonably well. And part of that modeling is that our eye scans around the perspective rendering using foveal sharpness to discern details much the same way as our scans around the real world using foveal sharpness to discern details.

The straight lines of rectilinear perspective renderings are usually not a problem because the model is usually good enough. Except if you're an ancient Greek building temples for the Gods, then [entasis](https://en.wikipedia.org/wiki/Entasis) comes into play...but you didn't have rectilinear perspective anyway.


[a]: Another example of this is that we discern color across the entire visual field despite color vision being largely limited to the fovea.

[b]: The station point of a perspective rendering is similar to the point of maximum cone density in the human eye.

  [1]: https://en.wikipedia.org/wiki/All_models_are_wrong