Integrated Videos and Maps for Driving Directions

This paper also has a webpage, which contains an overview video, as well as a link to the paper's full text. The demonstration starts around 3 minutes into the video.

In this paper, five researchers propose and implement a more effective system for giving driving directions. They point out that drivers are more comfortable driving a route the second time, because of the visual memory recalled from the first time. Their system, called Videomap, generates a video of a requested route (using previously captured panoramic images) and presents it to the user. Videomap uses several techniques to improve its usefulness:

• Portions of the route between turns are shown quickly, while turns are shown slowly
• The field of view is expanded near turns to take in (currently hand-selected) landmarks
• The video smooths out turns by rotating the field of view before the car itself turns
• Landmarks are freeze-framed alongside the video, while the video continues "driving"
  

After developing this tool, the researchers performed a user-study in which they gave users directions that they then had to follow. One group was given directions using Videomap, while the other used Picturemap, which only used pictures of the landmarks, not video. After viewing the directions in Videomap or Picturemap, the users then were presented with a real-time simulation of driving the route. Each group was also provided with a printed map of directions that they could use while "driving". At each intersection, the users were required to choose the correct path to take. Videomap was shown to be the better of the two applications in successful turns, times the user referenced the printed map, and user opinion.

Now, for some of my own thoughts on the paper. I found this paper to be very interesting, since it takes a novel approach to the common activity of getting directions. I believe that I would actually use such a system if it continued to be refined, and if some of the important caveats were handled well. I didn't really see any faults with this paper, except that the landmarks have to be hand-generated at the moment. I wouldn't be surprised if Google has their eye on this, especially since three of the researchers are with Microsoft! That being said, an obvious area for future work would be to hook up with a corporation like Microsoft or Google, and use their resources to improve and beta test this application.