Hacker News Reader

Comment by frik | Hacker News Reader

How does Google's Project Tango work anyway?

It's not LIDAR, nor Stereo-camera or Kinect 1 tech (infrared laser combined with a monochrome CMOS).

Does it use time-of-flight camera (like Kinect 2) or structure-from-motion (like MS Photosynth, ETH Zurich app)?

Tango uses this special processor (so structure-from-motion could be possible, as well as time-of-flight):

  It produces over 1 teraflop of processing power on only 
  a few hundred milliwatts of power

-- http://techcrunch.com/2014/02/20/inside-the-revolutionary-3d...

The Wikipedia article is still a stub: http://en.wikipedia.org/wiki/Project_Tango

View on HN

From the iFixit teardown: http://www.ifixit.com/Teardown/Project+Tango+Teardown/23835

Tango has a Primesense structured light system, a structure-from-motion coprocessor (or two) and a special depth sensing rear camera (IIRC).

Thanks! That makes sense.

So Google is using ETH's structure from motion technology (see comment below).

Hehe, google threw everything and a kitchen sink in there to cover all the bases. Even Primesense ")

Looks like Movidius started out as a mobile GPU in 2005/2008 :o ... and ended up being general purpose vector coprocessor / DSP

http://www.hotchips.org/wp-content/uploads/hc_archives/hc23/...

50Gflops/W in 65nm. Adapteva Epiphany http://www.parallella.org did 25 at 65nm, does 70Gflops/W in 28nm. Movidius projected 450Gflops/W at 28nm :o

With that great spec, why did google release tango with limited functionally, and not as a full blown processor ? And is it their next step ?

Full blown processors lead to full blown maintenance needs.

Im sure cost is also a big factor. Endgame is a small VERY CHEAP design that can be integrated with CMOS camera silicon.

Just like FLIR. They started with big expensive uncooled bolometers that required analog amplification (+1 chip), digitization (+1 chip) and processing (+1 more chip), and ended up producing all in one Lepton chip that interfaces using MIPI like every cellphone camera. They went from sensor + 2 analog chips + fpga to one small piece of silicon.

After all Google wants every android phone of the future to be able to do this. This means making every manufacturer buy additional hardware. It needs to be cheap and easy to use.

In the end Tango will probably be a licensed core that camera manufacturers can integrate into their own designs. One camera that does vision _and_ maps environment on the side.

What is the "ETH Zurich app", Please? :)

https://web.archive.org/web/20140103233938/http://www.inf.et... (ETH removed the website recently)

given that ETH is on Tango and that page also mentions SuiteSparse (also used for Google autonomous cars), it could be the case that Tango is based on Structure from Motion.

http://www.cise.ufl.edu/research/sparse/SuiteSparse/

another related page is still up: http://www.inf.ethz.ch/personal/acohen/papers/symmetryBA.php

Reply on news.ycombinator.com