[ulyssesgrant]

As someone who's worked on visual odometry projects before, I'm extremely interested in how they're able to compute the phone's motion estimations so accurately and in real time on a mobile device. Is any of this project open source? Is anyone here familiar with the vision techniques they're using to accomplish this feat? Perhaps there's an FPGA or ASIC designed for visual odometry inside the phone.

[siera]

It uses a Myriad 1 processor from Movidius http://www.movidius.com/tango. It's a custom media coprocessor aimed at mobile devices. I found more information about the processor architecture in these slides : http://www.hotchips.org/wp-content/uploads/hc_archives/hc23/...

Edit : Video of the talk associated with the slides http://www.youtube.com/watch?v=KiTaPRtN2yM

[ulyssesgrant]

Thanks for the info. Still very impressed with the speed and accuracy, assuming it works as well as they have presented. To be able to compute visual odometry in an arbitrary scene on a mobile device seems like a really big accomplishment, no?

[siera]

Visual odometry is getting quite accurate. For example, it is used on the Mars Exploration Rovers (Spirit and Opportunity) to improve the position obtained from wheel odometry [1]. As you mentionned, the big accomplishment is doing this in real time on a mobile device. On a typical mobile ARM processor, it would probably be very slow and drain the battery in no time. That's why they are using a custom coprocessor designed specifically for image processing and computer vision. The next challenge is porting the existing visual odometry algorithms on this new processor. It looks like hiDOF did the implementation [2].

[1] http://onlinelibrary.wiley.com/doi/10.1002/rob.20184/abstrac...

[2] http://hidof.com/projects/project-tango/