The Kinect 2.0 changes

[Gil]

Most of the information is from here:
http://video.wired.com/watch/new-xbox-kinect-exclusive-wired-video-398878

Time of flight camera, 1080p video, unknown depth resolution, wide
field of view, 6 person tracking, orientation tracking, muscle
tracking, force tracking, unknown audio capabilities.

[mankoff]

And more "official" info here: https://blogs.msdn.com/b/kinectforwindows/archive/2013/05/23/the-new-generation-kinect-for-windows-sensor-is-coming-next-year.aspx

more of the same above, but mention of all-new active-IR that will "allow the new sensor to work in nearly any lighting condition". Not sure what that means. The original Kinect was also active-IR.

   -k.

[Jeff Kramer]

The new sensor is ToF IR that they bought from Canesta (likely). 

The "in any lighting condition" means any lighting indoors, not outside. 

Sent from my iPhone

--
You received this message because you are subscribed to the Google Groups "OpenKinect" group.
To unsubscribe from this group and stop receiving emails from it, send an email to openkinect+...@googlegroups.com.
For more options, visit https://groups.google.com/groups/opt_out.
 
 

[Joshua Blake]

Jeff: Either or both Canesta or 3DV.

Ken: From the video walk-throughs, it appears the differences is that the "active IR" mode is a full flood of IR, which makes it suitable for use in further image processing, facial recognition in the dark, etc., while the current Kinect projects the IR dot patterns resulting in an IR image that cannot be used except for depth.

Based upon images of the back of the Xbox One:

http://www.wired.com/images_blogs/gadgetlab/2013/05/20130514-XBOX-ONE-013.jpg

http://2eq9hztv2wc1k6odx469m9znq0.wpengine.netdna-cdn.com/wp-content/uploads/back.jpg

The Kinect port (third from the right, marked with a Kinect silhouette above it) looks like a modified USB 3.0 B-type port. It is probably modified to carry extra power, similar to the Xbox 360-S orange USB Kinect connection. Various publications report (and the port markings match) the other USB connections are USB 3.0.

While the Xbox One Kinect 2 would not have a power brick adapter, it's likely that the Kinect for Windows model would have one.

Any reverse engineering efforts for open source Kinect 2 drivers would likely need to either use a K4W 2 sensor, which won't be available retail till next year per the K4W blog, or possibly hack up a USB 3 & power adapter for the Xbox Kinect 2 port.

Josh

---
Joshua Blake
Kinect for Windows MVP
OpenKinect Community Founder http://openkinect.org

(cell) 703-946-7176
Twitter: http://twitter.com/joshblake
Blog: http://nui.joshland.org

 

[Nicolas Burrus]

Well, I don't know about Kinect 2, but depending on the type of ToF
technology used, it could work outdoor, PMD cameras manage to do
fairly well. But they did not communicate about that so far AFAIK...

[Florian Echtler]

On 23.05.2013 20:43, Joshua Blake wrote:
> Ken: From the video walk-throughs, it appears the differences is that
> the "active IR" mode is a full flood of IR, which makes it suitable for
> use in further image processing, facial recognition in the dark, etc.,
> while the current Kinect projects the IR dot patterns resulting in an IR
> image that cannot be used except for depth.

Basically just the infrared camera image with the depth information removed?

> The Kinect port (third from the right, marked with a
> Kinect silhouette above it) looks like a modified USB 3.0 B-type port.
> It is probably modified to carry extra power, similar to the Xbox 360-S
> orange USB Kinect connection. Various publications report (and the port
> markings match) the other USB connections are USB 3.0.

USB 3.0 is specified for (at least) double the power of USB 2.0, so it's
perhaps just a non-standard plug.

> Any reverse engineering efforts for open source Kinect 2 drivers would
> likely need to either use a K4W 2 sensor, which won't be available
> retail till next year per the K4W blog, or possibly hack up a USB 3 &
> power adapter for the Xbox Kinect 2 port.

Any idea if it's also going to be sold separately (the Xbox version)?
Otherwise, I'll probably have to pre-order an Xbox now :-)

We'll probably need someone with access to an USB 3.0 protocol analyzer,
cost is still around 5000 US-$. Any takers?

Florian
--
SENT FROM MY DEC VT50 TERMINAL

[Joshua Blake]

> Basically just the infrared camera image with the depth information removed?

Not so much removed, just the raw data is more useful because of the way the ToF sensor works. The current Kinect can provide a raw IR image, which is literally what the IR sensor sees, and the depth image, which is a different image internally processed from the IR image. The raw IR image has bright IR dots like here:

http://www.cs.virginia.edu/~jfw3x/cs4501/images/IR_0.jpg

and the non-dots are very poorly illuminated, if at all. That IR image is not useful by itself for image processing, unless you turn off or block the IR projector and provide your own IR flood light, during which the depth data will not be available.

The new Kinect's ToF sensor sufficiently and evenly illuminates the scene with IR light such that the raw view is useful even while depth data is available:

http://cdn.thenextweb.com/wp-content/blogs.dir/1/files/2013/05/ces-3-2-520x347.jpg

> Any idea if it's also going to be sold separately (the Xbox version)?

I imagine if one can buy extra or replacement controllers, one could probably buy a replacement Kinect, but it may not be as available as extra controllers. The main difference is likely that if replacement Xbox One Kinects are available, they probably won't have a power adapter like the current generation stand-alone Xbox Kinects. The current ones have the power adapter because the old Xbox didn't have the special extra power orange USB port.

> We'll probably need someone with access to an USB 3.0 protocol analyzer,

Maybe only if you start with the Xbox One Kinect. We could possibly use a software sniffer/analyzer with the K4W 2 sensor.

Josh

---
Joshua Blake
Kinect for Windows MVP
OpenKinect Community Founder http://openkinect.org

(cell) 703-946-7176
Twitter: http://twitter.com/joshblake
Blog: http://nui.joshland.org

 

[Peter A]

If this is based off of Canesta's technology, I believe that it should have a global shutter too.  Which is very good news for robotics applications.  Their sensor I also think works outdoors.

- Peter

--
You received this message because you are subscribed to the Google Groups "OpenKinect" group.
To unsubscribe from this group and stop receiving emails from it, send an email to openkinect+...@googlegroups.com.
For more options, visit https://groups.google.com/groups/opt_out.
 
 

--
"Now, now my good man, this is no time for making enemies."    — Voltaire (1694-1778), on his deathbed in response to a priest asking that he renounce Satan.

[jeff kramer]

@Nicolas: How are the PMD cameras escaping the noise floor?

@Josh: You usually get the "raw view" for free - you're actually capturing 4 frames for each one that you see fully illuminated (and to calculate depth).

@Peter: Canesta was great indoors.  In 2008, their sensor wasn't good at all outside - couldn't escape the noise floor from the sun.  Maybe it is better now? =)

Jeff

[Nicolas Burrus]

The PMD cameras are measuring the phase-shift of a modulated IR signal
(the infrared light is emitted with a particular frequency), so on the
receiver side they can filter out any signal which is not modulated at
the exact same frequency, like the uniform background signal received
from the sun. Also, they only need to measure the phase of the
reflected signal, the absolute amplitude is not so important. There is
still more noise outdoor, but with the Camcube 3.0 we could still get
useful data.

I'm not sure about Canesta or 3DV technology though.

[Florian Echtler]

Nicolas Burrus <nicolas...@gmail.com> schrieb:

>The PMD cameras are measuring the phase-shift of a modulated IR signal
>(the infrared light is emitted with a particular frequency), so on the
>receiver side they can filter out any signal which is not modulated at
>the exact same frequency, like the uniform background signal received
>from the sun. Also, they only need to measure the phase of the
>reflected signal, the absolute amplitude is not so important. There is
>still more noise outdoor, but with the Camcube 3.0 we could still get
>useful data.

Out of curiosity, what's the modulation frequency? I'm assuming something in the GHz range?

Florian
--
SENT FROM MY PDP-11

[Peter A]

When I talked to Canesta several years ago, that's how theirs worked.  If you went to far away it would think the object was close.  They had some ideas on how to overcome that problem too.

For working outdoors, I just did a quick search on google scholar and found a few papers.    That technology has some limited success outdoors.  Here is one paper:

https://www2.lirmm.fr/lirmm/interne/BIBLI/CDROM/ROB/2007/ICRA%202007/data/papers/1361.pdf

- Peter

[nin...@gmail.com]

Any news if there is some type of automated hardware calibration or is this a factory setting / manual.

I would hope some genius decided to include an RGB pico projector for real world augmentation (project graphics onto bodies and objects held in hands for clothes, blood, targets, interactive white boards etc)
Sent wirelessly from my BlackBerry device on the Bell network.
Envoyé sans fil par mon terminal mobile BlackBerry sur le réseau de Bell.

[Jeff Kramer]

Thanks Nicolas and Peter!

I hadn't thought that amplitude wasn't as important. I think that, depending on price, this could be a really valuable addition. 

Jeff

Sent from my iPhone

[Jeff Kramer]

Actually MHz. The higher the frequency, the shorter the wavelength, so the shorter "wraparound" distance you get. SR4000 run at 15 MHz for 10 meter range and 30 MHz for 5 meter range.

Jeff

Sent from my iPhone

[Lorne Covington]

This is the main problem I have with the Intel Creative Gesture Camera; it "aliases" the depth data.� It's designed for near use, but it's apparently got a pretty bright IR source so objects past it's outer distance fold back into the depth scene - so something at 3 meters looks like it's at 1 meter.� This foiled my attempts to use it to generate point-cloud data.

I'm totally guessing, but that would seem to be a problem with any ToF camera that did not use very short pulsed light for illumination.� Seems like you could use another sensor to measure relative amplitude to help mask it, but not sure it could be done reliably.

I'm guessing the new Kinect deals with this by just having a large(ish) native capture range so it's unlikely to get aliasing in the average living room.� Could be a problem for applications in more open areas (the kinds of places I would like to use it of course).

Time will tell!

- Lorne

http://noirflux.com


On 5/24/2013 8:25 AM, Peter A wrote:

When I talked to Canesta several years ago, that's how theirs worked.� If you went to far away it would think the object was close.� They had some ideas on how to overcome that problem too.

For working outdoors, I just did a quick search on google scholar and found a few papers.��� That technology has some limited success outdoors.� Here is one paper:

https://www2.lirmm.fr/lirmm/interne/BIBLI/CDROM/ROB/2007/ICRA%202007/data/papers/1361.pdf

- Peter

On Fri, May 24, 2013 at 2:57 AM, Nicolas Burrus <nicolas...@gmail.com> wrote:

The PMD cameras are measuring the phase-shift of a modulated IR signal
(the infrared light is emitted with a particular frequency), so on the
receiver side they can filter out any signal which is not modulated at
the exact same frequency, like the uniform background signal received
from the sun. Also, they only need to measure the phase of the
reflected signal, the absolute amplitude is not so important. There is
still more noise outdoor, but with the Camcube 3.0 we could still get
useful data.

I'm not sure about Canesta or 3DV technology though.

On Thu, May 23, 2013 at 11:01 PM, jeff kramer <jeff...@gmail.com> wrote:
> @Nicolas: How are the PMD cameras escaping the noise floor?
> @Josh: You usually get the "raw view" for free - you're actually capturing 4
> frames for each one that you see fully illuminated (and to calculate depth).

> @Peter: Canesta was great indoors. �In 2008, their sensor wasn't good at all
> outside - couldn't escape the noise floor from the sun. �Maybe it is better

> now? =)
>
> Jeff
>
>
> On Thu, May 23, 2013 at 3:58 PM, Peter A <peter....@gmail.com> wrote:
>>
>> If this is based off of Canesta's technology, I believe that it should

>> have a global shutter too. �Which is very good news for robotics
>> applications. �Their sensor I also think works outdoors.

>> "Now, now my good man, this is no time for making enemies." � �� Voltaire

>> (1694-1778), on his deathbed in response to a priest asking that he renounce
>> Satan.
>>
>> --
>> You received this message because you are subscribed to the Google Groups
>> "OpenKinect" group.
>> To unsubscribe from this group and stop receiving emails from it, send an
>> email to openkinect+...@googlegroups.com.
>> For more options, visit https://groups.google.com/groups/opt_out.
>>
>>
>
>
> --
> You received this message because you are subscribed to the Google Groups
> "OpenKinect" group.
> To unsubscribe from this group and stop receiving emails from it, send an
> email to openkinect+...@googlegroups.com.
> For more options, visit https://groups.google.com/groups/opt_out.
>
>

--
You received this message because you are subscribed to the Google Groups "OpenKinect" group.
To unsubscribe from this group and stop receiving emails from it, send an email to openkinect+...@googlegroups.com.
For more options, visit https://groups.google.com/groups/opt_out.

--
"Now, now my good man, this is no time for making enemies."� � � Voltaire (1694-1778), on his deathbed in response to a priest asking that he renounce Satan.

--
You received this message because you are subscribed to the Google Groups "OpenKinect" group.
To unsubscribe from this group and stop receiving emails from it, send an email to openkinect+...@googlegroups.com.
For more options, visit https://groups.google.com/groups/opt_out.

�
�

-- 

http://noirflux.com

[Björn Giesler]

Hi,

when we worked with PMD, we used more than one frequency to solve this problem and got theoretical ranges of up to a couple hundred meters before folding back to zero (although the light sources used did not allow realistic measurements of anything farther away than 90m or so). So the problem is solvable, although obviously you would need precise control over the modulation rate to do this, which is probably not possible with Kinect. 

But even one modulation frequency will be quite enough. 4Mhz gives about 37.5m of unambiguity, which should be much more than the light source built into the Kinect can provide. 

Regards

Björn

Von meinem iPhone gesendet

Am 24.05.2013 um 15:58 schrieb Lorne Covington <mediado...@gmail.com>:

This is the main problem I have with the Intel Creative Gesture Camera; it "aliases" the depth data.  It's designed for near use, but it's apparently got a pretty bright IR source so objects past it's outer distance fold back into the depth scene - so something at 3 meters looks like it's at 1 meter.  This foiled my attempts to use it to generate point-cloud data.

I'm totally guessing, but that would seem to be a problem with any ToF camera that did not use very short pulsed light for illumination.  Seems like you could use another sensor to measure relative amplitude to help mask it, but not sure it could be done reliably.

I'm guessing the new Kinect deals with this by just having a large(ish) native capture range so it's unlikely to get aliasing in the average living room.  Could be a problem for applications in more open areas (the kinds of places I would like to use it of course).

Time will tell!

- Lorne

http://noirflux.com


On 5/24/2013 8:25 AM, Peter A wrote:

When I talked to Canesta several years ago, that's how theirs worked.  If you went to far away it would think the object was close.  They had some ideas on how to overcome that problem too.

For working outdoors, I just did a quick search on google scholar and found a few papers.    That technology has some limited success outdoors.  Here is one paper:

https://www2.lirmm.fr/lirmm/interne/BIBLI/CDROM/ROB/2007/ICRA%202007/data/papers/1361.pdf

- Peter

On Fri, May 24, 2013 at 2:57 AM, Nicolas Burrus <nicolas...@gmail.com> wrote:

The PMD cameras are measuring the phase-shift of a modulated IR signal
(the infrared light is emitted with a particular frequency), so on the
receiver side they can filter out any signal which is not modulated at
the exact same frequency, like the uniform background signal received
from the sun. Also, they only need to measure the phase of the
reflected signal, the absolute amplitude is not so important. There is
still more noise outdoor, but with the Camcube 3.0 we could still get
useful data.

I'm not sure about Canesta or 3DV technology though.

On Thu, May 23, 2013 at 11:01 PM, jeff kramer <jeff...@gmail.com> wrote:
> @Nicolas: How are the PMD cameras escaping the noise floor?
> @Josh: You usually get the "raw view" for free - you're actually capturing 4
> frames for each one that you see fully illuminated (and to calculate depth).

> @Peter: Canesta was great indoors.  In 2008, their sensor wasn't good at all
> outside - couldn't escape the noise floor from the sun.  Maybe it is better

> now? =)
>
> Jeff
>
>
> On Thu, May 23, 2013 at 3:58 PM, Peter A <peter....@gmail.com> wrote:
>>
>> If this is based off of Canesta's technology, I believe that it should

>> have a global shutter too.  Which is very good news for robotics
>> applications.  Their sensor I also think works outdoors.

>> "Now, now my good man, this is no time for making enemies."    — Voltaire

>> (1694-1778), on his deathbed in response to a priest asking that he renounce
>> Satan.
>>
>> --
>> You received this message because you are subscribed to the Google Groups
>> "OpenKinect" group.
>> To unsubscribe from this group and stop receiving emails from it, send an
>> email to openkinect+...@googlegroups.com.
>> For more options, visit https://groups.google.com/groups/opt_out.
>>
>>
>
>
> --
> You received this message because you are subscribed to the Google Groups
> "OpenKinect" group.
> To unsubscribe from this group and stop receiving emails from it, send an
> email to openkinect+...@googlegroups.com.
> For more options, visit https://groups.google.com/groups/opt_out.
>
>

--
You received this message because you are subscribed to the Google Groups "OpenKinect" group.
To unsubscribe from this group and stop receiving emails from it, send an email to openkinect+...@googlegroups.com.
For more options, visit https://groups.google.com/groups/opt_out.

--
"Now, now my good man, this is no time for making enemies."    — Voltaire (1694-1778), on his deathbed in response to a priest asking that he renounce Satan.

--
You received this message because you are subscribed to the Google Groups "OpenKinect" group.
To unsubscribe from this group and stop receiving emails from it, send an email to openkinect+...@googlegroups.com.
For more options, visit https://groups.google.com/groups/opt_out.
 
 

-- 

http://noirflux.com

[Lorne Covington]

On 5/24/2013 11:06 AM, Bj�rn Giesler wrote:

Hi,

when we worked with PMD, we used more than one frequency to solve this problem and got theoretical ranges of up to a couple hundred meters before folding back to zero (although the light sources used did not allow realistic measurements of anything farther away than 90m or so). So the problem is solvable, although obviously you would need precise control over the modulation rate to do this, which is probably not possible with Kinect.

Gee only 90m!� Yow!

But even one modulation frequency will be quite enough. 4Mhz gives about 37.5m of unambiguity, which should be much more than the light source built into the Kinect can provide.

Is depth resolution always proportional to the max distance, since you're always just measuring a phase difference?� So by going for greater range you're sacrificing near-field accuracy?� If so I could see why the Intel cam aliases at such close distances as it is designed for in-front-of-a-screen interactions.

Thanks Bj�rn!

- Lorne

Regards

Bj�rn

Von meinem iPhone gesendet

Am 24.05.2013 um 15:58 schrieb Lorne Covington <mediado...@gmail.com>:

This is the main problem I have with the Intel Creative Gesture Camera; it "aliases" the depth data.� It's designed for near use, but it's apparently got a pretty bright IR source so objects past it's outer distance fold back into the depth scene - so something at 3 meters looks like it's at 1 meter.� This foiled my attempts to use it to generate point-cloud data.

I'm totally guessing, but that would seem to be a problem with any ToF camera that did not use very short pulsed light for illumination.� Seems like you could use another sensor to measure relative amplitude to help mask it, but not sure it could be done reliably.

I'm guessing the new Kinect deals with this by just having a large(ish) native capture range so it's unlikely to get aliasing in the average living room.� Could be a problem for applications in more open areas (the kinds of places I would like to use it of course).

Time will tell!

- Lorne

http://noirflux.com


On 5/24/2013 8:25 AM, Peter A wrote:

When I talked to Canesta several years ago, that's how theirs worked.� If you went to far away it would think the object was close.� They had some ideas on how to overcome that problem too.

For working outdoors, I just did a quick search on google scholar and found a few papers.��� That technology has some limited success outdoors.� Here is one paper:

https://www2.lirmm.fr/lirmm/interne/BIBLI/CDROM/ROB/2007/ICRA%202007/data/papers/1361.pdf

- Peter

On Fri, May 24, 2013 at 2:57 AM, Nicolas Burrus <nicolas...@gmail.com> wrote:

The PMD cameras are measuring the phase-shift of a modulated IR signal
(the infrared light is emitted with a particular frequency), so on the
receiver side they can filter out any signal which is not modulated at
the exact same frequency, like the uniform background signal received
from the sun. Also, they only need to measure the phase of the
reflected signal, the absolute amplitude is not so important. There is
still more noise outdoor, but with the Camcube 3.0 we could still get
useful data.

I'm not sure about Canesta or 3DV technology though.

On Thu, May 23, 2013 at 11:01 PM, jeff kramer <jeff...@gmail.com> wrote:
> @Nicolas: How are the PMD cameras escaping the noise floor?
> @Josh: You usually get the "raw view" for free - you're actually capturing 4
> frames for each one that you see fully illuminated (and to calculate depth).

> @Peter: Canesta was great indoors. �In 2008, their sensor wasn't good at all
> outside - couldn't escape the noise floor from the sun. �Maybe it is better

> now? =)
>
> Jeff
>
>
> On Thu, May 23, 2013 at 3:58 PM, Peter A <peter....@gmail.com> wrote:
>>
>> If this is based off of Canesta's technology, I believe that it should

>> have a global shutter too. �Which is very good news for robotics
>> applications. �Their sensor I also think works outdoors.

>> "Now, now my good man, this is no time for making enemies." � �� Voltaire

>> (1694-1778), on his deathbed in response to a priest asking that he renounce
>> Satan.
>>
>> --
>> You received this message because you are subscribed to the Google Groups
>> "OpenKinect" group.
>> To unsubscribe from this group and stop receiving emails from it, send an
>> email to openkinect+...@googlegroups.com.
>> For more options, visit https://groups.google.com/groups/opt_out.
>>
>>
>
>
> --
> You received this message because you are subscribed to the Google Groups
> "OpenKinect" group.
> To unsubscribe from this group and stop receiving emails from it, send an
> email to openkinect+...@googlegroups.com.
> For more options, visit https://groups.google.com/groups/opt_out.
>
>

--
You received this message because you are subscribed to the Google Groups "OpenKinect" group.
To unsubscribe from this group and stop receiving emails from it, send an email to openkinect+...@googlegroups.com.
For more options, visit https://groups.google.com/groups/opt_out.

--
"Now, now my good man, this is no time for making enemies."� � � Voltaire (1694-1778), on his deathbed in response to a priest asking that he renounce Satan.

--
You received this message because you are subscribed to the Google Groups "OpenKinect" group.
To unsubscribe from this group and stop receiving emails from it, send an email to openkinect+...@googlegroups.com.
For more options, visit https://groups.google.com/groups/opt_out.

�
�

-- 

http://noirflux.com

--
You received this message because you are subscribed to the Google Groups "OpenKinect" group.
To unsubscribe from this group and stop receiving emails from it, send an email to openkinect+...@googlegroups.com.
For more options, visit https://groups.google.com/groups/opt_out.

�
�

--
You received this message because you are subscribed to the Google Groups "OpenKinect" group.
To unsubscribe from this group and stop receiving emails from it, send an email to openkinect+...@googlegroups.com.
For more options, visit https://groups.google.com/groups/opt_out.

�
�

[jeff kramer]

I think depth resolution suffers less than integration time.  While you can build a very sensitive beat detection circuit, for each "depth image" you have to capture 4 regular images at different illumination levels.

-Jeff

On Fri, May 24, 2013 at 12:24 PM, Lorne Covington <mediado...@gmail.com> wrote:

On 5/24/2013 11:06 AM, Björn Giesler wrote:

Hi,

when we worked with PMD, we used more than one frequency to solve this problem and got theoretical ranges of up to a couple hundred meters before folding back to zero (although the light sources used did not allow realistic measurements of anything farther away than 90m or so). So the problem is solvable, although obviously you would need precise control over the modulation rate to do this, which is probably not possible with Kinect.

Gee only 90m!  Yow!

But even one modulation frequency will be quite enough. 4Mhz gives about 37.5m of unambiguity, which should be much more than the light source built into the Kinect can provide.

Is depth resolution always proportional to the max distance, since you're always just measuring a phase difference?  So by going for greater range you're sacrificing near-field accuracy?  If so I could see why the Intel cam aliases at such close distances as it is designed for in-front-of-a-screen interactions.

Thanks Björn!

- Lorne

Regards

Björn

Von meinem iPhone gesendet

Am 24.05.2013 um 15:58 schrieb Lorne Covington <mediado...@gmail.com>:

This is the main problem I have with the Intel Creative Gesture Camera; it "aliases" the depth data.  It's designed for near use, but it's apparently got a pretty bright IR source so objects past it's outer distance fold back into the depth scene - so something at 3 meters looks like it's at 1 meter.  This foiled my attempts to use it to generate point-cloud data.

I'm totally guessing, but that would seem to be a problem with any ToF camera that did not use very short pulsed light for illumination.  Seems like you could use another sensor to measure relative amplitude to help mask it, but not sure it could be done reliably.

I'm guessing the new Kinect deals with this by just having a large(ish) native capture range so it's unlikely to get aliasing in the average living room.  Could be a problem for applications in more open areas (the kinds of places I would like to use it of course).

Time will tell!

- Lorne

http://noirflux.com


On 5/24/2013 8:25 AM, Peter A wrote:

When I talked to Canesta several years ago, that's how theirs worked.  If you went to far away it would think the object was close.  They had some ideas on how to overcome that problem too.

For working outdoors, I just did a quick search on google scholar and found a few papers.    That technology has some limited success outdoors.  Here is one paper:

https://www2.lirmm.fr/lirmm/interne/BIBLI/CDROM/ROB/2007/ICRA%202007/data/papers/1361.pdf

- Peter

On Fri, May 24, 2013 at 2:57 AM, Nicolas Burrus <nicolas...@gmail.com> wrote:

The PMD cameras are measuring the phase-shift of a modulated IR signal
(the infrared light is emitted with a particular frequency), so on the
receiver side they can filter out any signal which is not modulated at
the exact same frequency, like the uniform background signal received
from the sun. Also, they only need to measure the phase of the
reflected signal, the absolute amplitude is not so important. There is
still more noise outdoor, but with the Camcube 3.0 we could still get
useful data.

I'm not sure about Canesta or 3DV technology though.

On Thu, May 23, 2013 at 11:01 PM, jeff kramer <jeff...@gmail.com> wrote:
> @Nicolas: How are the PMD cameras escaping the noise floor?
> @Josh: You usually get the "raw view" for free - you're actually capturing 4
> frames for each one that you see fully illuminated (and to calculate depth).

> @Peter: Canesta was great indoors.  In 2008, their sensor wasn't good at all
> outside - couldn't escape the noise floor from the sun.  Maybe it is better

> now? =)
>
> Jeff
>
>
> On Thu, May 23, 2013 at 3:58 PM, Peter A <peter....@gmail.com> wrote:
>>
>> If this is based off of Canesta's technology, I believe that it should

>> have a global shutter too.  Which is very good news for robotics
>> applications.  Their sensor I also think works outdoors.

>> "Now, now my good man, this is no time for making enemies."    — Voltaire

>> (1694-1778), on his deathbed in response to a priest asking that he renounce
>> Satan.
>>
>> --
>> You received this message because you are subscribed to the Google Groups
>> "OpenKinect" group.
>> To unsubscribe from this group and stop receiving emails from it, send an
>> email to openkinect+...@googlegroups.com.
>> For more options, visit https://groups.google.com/groups/opt_out.
>>
>>
>
>
> --
> You received this message because you are subscribed to the Google Groups
> "OpenKinect" group.
> To unsubscribe from this group and stop receiving emails from it, send an
> email to openkinect+...@googlegroups.com.
> For more options, visit https://groups.google.com/groups/opt_out.
>
>

--
You received this message because you are subscribed to the Google Groups "OpenKinect" group.
To unsubscribe from this group and stop receiving emails from it, send an email to openkinect+...@googlegroups.com.
For more options, visit https://groups.google.com/groups/opt_out.

--
"Now, now my good man, this is no time for making enemies."    — Voltaire (1694-1778), on his deathbed in response to a priest asking that he renounce Satan.

--
You received this message because you are subscribed to the Google Groups "OpenKinect" group.
To unsubscribe from this group and stop receiving emails from it, send an email to openkinect+...@googlegroups.com.
For more options, visit https://groups.google.com/groups/opt_out.
 
 

-- 

http://noirflux.com

--
You received this message because you are subscribed to the Google Groups "OpenKinect" group.
To unsubscribe from this group and stop receiving emails from it, send an email to openkinect+...@googlegroups.com.
For more options, visit https://groups.google.com/groups/opt_out.
 
 

--

You received this message because you are subscribed to the Google Groups "OpenKinect" group.
To unsubscribe from this group and stop receiving emails from it, send an email to openkinect+...@googlegroups.com.
For more options, visit https://groups.google.com/groups/opt_out.
 
 

[Björn Giesler]

Depth accuracy is linearly dependant on modulation frequency, but also a lot of other factors. We had roughly 0,3% of accuracy (10cm of accuracy at 30m) on pure-white objects, IIRC. So if the Intel camera is not crazy accurate, its small range probably has other reasons…

…and 90m is not that much for Automotive purposes...

Jeff, why would you have to capture 4 regular images? Technologies are diverse in this field, PMD works wildly different than Canesta does, but 4 images at different illumination levels I don't get…

Björn

Björn Giesler
bjo...@giesler.de
Verschlüssel Deine Emails -- oder schreibst Du Deine Briefe außen auf den Umschlag?
Meinen Schlüssel findest Du auf http://giesler.no-ip.org/~bjoern/downloads/bg.asc -- schick mir Deinen!
Mehr Informationen findest Du auf http://www.gnupg.org/index.de.html

[jeff kramer]

Capture 4 parts of the phase and integrate them - that's how the MESA SR4000 works.  "Regular images" was a bad way of stating it - you capture/compare at 4 different places (which CAN be parsed as a visible image).

So you capture the sine wave at the neg to pos crossing point, maximum, pos to neg crossing point and minimum.  You then integrate the differences in order to cut out some of the noise, etc.

Now, I just learned that PMD is on Audis - I'm VERY interested in seeing how they did that/use it outside. =)

[Björn Giesler]

Ah.

PMD works differently -- you modulate with a square wave, the camera cross-correlates the outgoing and incoming square waves on the chip itself, subtracting background light as "DC offset" (if you will call it that), and returns four equidistant values on the cross-correlation result, which is a triangle. atan((v1-v3)/(v2-v4)) gives phase, which together with mod. frequency gives range.

Regards,
Björn

[Nicolas Burrus]

You may be interested by this report then
http://cdn.intechopen.com/pdfs/5767/InTech-A_performance_review_of_3d_tof_vision_systems_in_comparison_to_stereo_vision_systems.pdf

The secret was more illumination units at that time :)

[jeff kramer]

Nicolas to the rescue.  Thanks man! =)

-Jeff

[Björn Giesler]

There you go. All explained, and this is the camera we were using.