ftw!
Home Blog Archives RSS
¶ photo-guided driving directions
An idea I think is cool, but will probably never get around to doing myself, so I'm throwing it out here.

One of the things we've been toying with at lab is the idea of photograph-guided navigation. The basic problem is that written driving directions are not optimal. Consider:
  • Turn Right at Main St. 0.5 miles
  • Turn Left at Park Ave. 1.4 miles
  • Bear Left at Broadway 0.3 miles
  • Arrive at Destination
Being able to follow these directions requires being able to identify "Main St", "Park Ave", and "Broadway". How many times have we tried to follow directions like these only to find that the street signs are difficult to see or not even there at all? The visual cue that we're given is unreliable. What if, instead of a bunch of street names, you were given a series of photographs?



Would that not be so much easier to follow than a little line that says "Turn right at Main Street". You could still have the text, of course, you'd just supplement it with a picture.

So how would something like this work? There are two things that need to happen. First, you'd need an image database of every intersection and point of interest, along with the location of the image and a direction vector. You need the direction vector because at any given point, you can have many images facing different directions that look completely different. It would be silly to give directions that say "Turn left when you see this picture behind you". Collecting all the images is easy. Take a laptop with a GPS unit, a webcam, and a big hard drive. The GPS unit provides accurate positioning data along with direction vectors. Point the camera forward at all times and keep it turned on. As the car drives, the laptop automatically grabs images from the webcam and stores them onto a local database that gets merged with a central database later on. A skilled programmer could write the program for this in a single day. This is actually how MapQuest built its street maps. They hired a bunch of people, gave them laptops and GPS units, and told them to drive around and type in all the street information they could find.

Second, you'd need to integrate this into a search engine. Google Maps seems like a good candidate, given that it's fairly easy to customize and integrate into your other web applications. When you get your list of directions from Google Maps, you compute a position and direction that you want a picture for and then look for an image in your database that matches up. A nice thing about it is that you could roll it out incrementally. If you don't have a picture available for an interesction, big deal, just keep showing what you've always shown before. Another thing you could do is to have two image classes - one for nighttime driving and one for daytime driving. The picture above shows an arrow overlaid on the image, pointing in the direction you want to turn. This arrow is easily calculated based on the angle of the turn (the same way they get the "Bear Right", "Slight Right", "Right" phrases.)

So anybody at Google read this blog? =P



No comments, be the first!




Comments disabled until the spammers go away. I hope you comment spammers all die horrible deaths and are forced to delete endless streams of comment spam in your days in purgatory.
• Powered by bBlog