Most of us who've used Photosynth have tried to synth a car at one time or another. Indeed, there are about 1,500 synths with the "car" tag already.
But if you've tried, you've probably found that car synths often don't work out as well as room synths, art synths, landscape synths, or even people synths. Let's discuss why that is, and what can be done about it.
One of the first steps in the Photosynth processing pipeline is "feature extraction". This is where we analyze each photo and look for areas of interesting texture that we hope can later be matched to textures in other photos. Notice that I said "texture", because Photosynth doesn't look for corners or edges; it looks for areas of visual "roughness". Our feature detection algorithm was inspired by the popular SIFT Algorithm created by David Lowe at UBC.
The problem with cars is that they have very little "visual roughness". Most cars are smooth and shiny, and that leads to very few matchable features. In fact, because if this, we've always said that dirty cars synth better than clean ones. :)
What often happens when synthing a car is that the background gets matched much more than the car, resulting in a very poor point cloud, and a frustrating navigation experience:
It looks like the car got matched, but when you turn off the photos and examine the point cloud...
... only the bushes in the background and the ground on the lower left got matched. Apart from a small patch on the rear door, the car itself was almost invisible to Photosynth. This makes for very poor connectivity and very poor navigation.
Nonetheless, if you pay attention to the following you CAN get some pretty good results. In increasing order of difficulty, Here's what you can try:
How to Shoot a Car Interior (easy)
How to Shoot a Car Exterior (hard)
How to Shoot a Booth at an Auto Show (ambitious!)