Taking the longer way for fun.
People act like time is always the most important factor on the journey from point A to B, when in fact it's enjoyability. In some cases, enjoyability is maximized by the lowest journey time because traffic is atrocious, more time can be spent at the destination, etc. In other cases, safety is the largest contributor: people would prefer an extra hour if it meant improving their survival odds by a decent amount. And in other-other cases, the raw enjoyability itself is the largest contributor, i.e., the journey is enjoyable because the journey itself. This enjoyment can come from the views, smells, sounds, memories and conversations conjured by something on the route.
And yet no navigation app offers a scenic route option.
Automatically quantifying scenicness is arguably challenging, but some rough attempts can be made. A segment is hereby defined as the stretch of route between intersections.
A street view image that is primarily blue or green may mean there is a lot of water, sky, or greenery, all of which are generally associated with scenicness.
Pixel color variation is likely not a good indicator.
Higher elevations relative to surrounding areas offer better vantage points, thus are generally more scenic than their lower counterparts. Of course, being able to actually see around said high area matters, so if trees are blocking the view, it doesn't matter.
Run a sentiment analysis on reviews taken throughout the segment.
See how many pictures have been taken along the route. If clusters exist in a single location, it's likely there is something "scenic" there.
See how many Places (capitalized, as in Google Maps Place (of interest)) exist on the segment. Scenic routes allow stopping to check things out. Places would be weighted on review score and number of reviews.
Train a model on what's scenic and what's not. This can be broken up into a few training categories (city, mountains, beaches, etc.) for a variety of routes and variation within routes.
Once the model is accurate, street view images would be fed to to take a five-point average of the front, left, right, back, and up views' scenicness. If there are positive outliers, weight them more heavily since we are looking for maximum scenicness in a view. For lengthy segments, perform X number of five-point averages, where X is determined by some equation I have no idea of, then average again, weighting any outliers accordingly.
A map of all segments would be created along with their associated travel times. All possible segments' scenicnesses and times that make up a route would be calculated, then the best ones chosen that fit the constraints. (I'm not familiar with optimization methods, so there may be an easier way than calculating every single route.)
If my journey is a three-hour drive, I don't want an eight-hour scenic drive, but maybe I want a four-hour one.
Users would be allowed to enter maximum added time to put an upper bound on the total trip length.
A maximum elevation gain would prevent more-than-desired physical activity and be used almost entirely for walking and cycling routes.
Multiple route options are offered, each detailing time and places of interest along the route. Users can select accordingly.