Saturday, October 30, 2010

Microsoft is Buying Canesta

Canesta announced that it has signed a definitive agreement to have its products, technology, intellectual property, customer contracts, and other resources acquired by the Microsoft.

No details of the agreement have been disclosed. The acquisition is expected to be completed before the end of this year.

Canesta has about 70 employees and has raised approximately $70 million. It has 44 patents granted to date and dozens more on file. There is a nice company fact sheet here.

Canesta team complements 3DV team acquired by Microsoft in June 2009. This shows that Microsoft internal R&D efforts continue to be concentrated on ToF technology. So, it looks like it's going to be a nice competition between the internally designed ToF sensor and Primesense's structured light - active stereo approach.

Primesense won over 3DV in the first generation Kinect. Do the combined 3DV and Canesta efforts give them a win in the next turn? - only time can tell.

Updates: The Street says it got an email confirmation of the deal from Microsoft saying "Microsoft has long pursued a vision of natural user interface. Canesta has developed some interesting technology for sensing gestures that complements advances already underway at Microsoft."

NYT quotes Steve Ballmer, Microsoft CEO, discussing the company’s plans to advance the gesture technology "well beyond video games".


  1. Congrats to Jim S., Cyrus B and Pat OC and the rest of the team at Canesta!

  2. Primesense investors must be feeling a lot of remorse right now.

  3. it's amusing to file such kind of patent:7,464,351!!!!
    We, all the CMOS sensor community, can be sued by Canesta! It's a real gabage patent!

  4. Probably '351 would not stand up to reexamination. I don't think anyone has anything to worry about from that patent. Looks like PTO required them to remove some claims from the parent app and file this one with the same specification. It is well known in the sensor community that fab rules were meant to be broken.

  5. but having such idea to patent this is amazing and crzay!

  6. The whole patent seems to be completely disconnected from its claims and abstract. I agree with Eric, it looks like a waste of attorney's time and Pixim money to me.

  7. The entire approach- TOF- is inherently deficient for any number of reason. I am guessing that they bought Canesta to avoid a troll scenario since Canesta's model is capital-intensive and long term. Re- PrimeSense. My guess is that they had committed to 3DVS before they agreed to PrimeSense. PrimeSense is a stop gap between now and whenever they launch an exotic pixel TOF solution. In the end, an active light source is not the answer.

  8. Canesta has been around for many years with only meager sales. After all the money in and all the money out, this deal is probably a net loss for the investors, or at least the original investors. Even a stellar organization like Photobit didn't do well at the acquisition.

  9. Yip. They would have had to be acquired for 400M+ for them to justify the investment (and they have had many many rounds of it!). Same thing happened with 3DVS (who basically got chopped up into little pieces and auctioned off). The problem is that this poisoned the waters for a number of startups looking to get VC funding, when the two TOF companies failed miserably. The problem is endemic of TOF as a technology. The answer to this technology, as MSFT will come to find out after much work is in stereo imaging with standard CMOS.

  10. the key point of ToF sensing is to capture few useful photons in a lot of ambiant "noise" photons. The two possible solutions are either using highly time concentrated laser impulse with ultra-fast detector or using long temporel integration on many laser/optical cycles. 3DV used a gated image intensified tube for the demo. This is a viable solution but definitively not low cost at all. Mesa, PMD uses demodulating based solution by integrating many optical pulses. The latence could be a problem too. Finally for PrimeSense, the structured light stereo solution makes sens, the only problem is that when the gaming guys find out that there is a X00mW laser inside, they will not very happy with.

    My questions are:
    1. is 3D necessary?
    2. can we play all the games with only two hands?
    3. what is really wrong with a remote controller?


  11. 1. 3D is necessary to identify ROIs with varying degrees of freedom. If you are tracking a face, the 2D is plenty, but hands change so much, because of the degrees of freedom associated with them. It makes more sense to isolate regions of interest based on 3D.

    2. No.

    3. Nothing. It's just very limiting when you are combining IPTV applications. You can utilize pointing devices, but ultimately, losing the device all together makes more sense and is more cost-effective.

    TOF and structured light are both bad. Light blasting the FOV is not a good idea.

  12. Is spad going to be a viable, low-cost solution?
    Considering it can be done with standard CMOS ?

  13. the QE in NIR is low. I guess that you have to fire your laser for each Z plane, so if the precision requirement is high, the laser power is limited too.

  14. a remote controller is much less expensive than 3D cam.

  15. Anonymous but close byNovember 5, 2010 at 6:01 AM

    Stereo imaging with CMOS sensors, lol. That approach was flushed long time ago, it's not even a joke anymore. The speculation here is comical. 3D input is not about replacing controllers.. that's rigid thinking. 3D input is about completely new user experiences, not about controller vs none.

  16. Stereo imaging with CMOS sensors is comical? What other stereo imaging is there? Are you referring to CCD sensors? You may wish to clarify. Your comment makes no sense. And who flushed that approach? I am guessing that you mean flushed out? By long time ago, do you mean to say many years ago, i.e. when compute power was significantly less and you could barely get flops/dollar/watt to a reasonable number?


All comments are moderated to avoid spam.