Field of View (Horizontal, Vertical, Diagonal) = 58° H, 45°
V, 70° D
Spatial x/y resolution (@ 2m distance from sensor) = 3mm
Depth z resolution (@ 2m distance from sensor) = 1cm
Operation range =
0.8m - 3.5m
Questions:
1) Do those above specs hold for the Kinect as well? (Since Kinect is
also using PS1080)
2) The stipulated X/Y resolution is 3mm and Z resolution 1cm. But from
the other posts in this forum, it seems like the accuracy is to 1mm
for all X/Y/Z. OpenNI programs also output values to the mm. Does
anyone know why the numbers are different?
3) Is the accuracy up to 3mm @2m for X/Y (similarly up to 1cm @2m for
Z), and the accuracy decreases when you move further away from the
sensor?
4) The operation range is stipulated at 0.8m - 3.5m, but from another
post, it says the Kinect output depth values at 30cm-10m. Why is there
a difference?
It would be good if someone can give a more technical explanation on
this please. Thank you!
> Field of View (Horizontal, Vertical, Diagonal) = 58° H, 45°
> V, 70° D
> Spatial x/y resolution (@ 2m distance from sensor) = 3mm
> Depth z resolution (@ 2m distance from sensor) = 1cm
> Operation range =
> 0.8m - 3.5m
> Questions:
> 1) Do those above specs hold for the Kinect as well? (Since Kinect is
> also using PS1080)
> 2) The stipulated X/Y resolution is 3mm and Z resolution 1cm. But from
> the other posts in this forum, it seems like the accuracy is to 1mm
> for all X/Y/Z. OpenNI programs also output values to the mm. Does
> anyone know why the numbers are different?
> 3) Is the accuracy up to 3mm @2m for X/Y (similarly up to 1cm @2m for
> Z), and the accuracy decreases when you move further away from the
> sensor?
> 4) The operation range is stipulated at 0.8m - 3.5m, but from another
> post, it says the Kinect output depth values at 30cm-10m. Why is there
> a difference?
> It would be good if someone can give a more technical explanation on
> this please. Thank you!
> The numbers for the kinect are more like 20mm X/Y resolution. > Microsoft significantly downscaled the resolution of the device to > make it cheaper.
> The reference implementation would have a much better resolution.
The Microsoft Kinect is basically the same sensor as the PrimeSense PS1080 using the same cameras and same chipset, with a "crippled" firmware (certain features have been taken out), so the two sensors produce exactly the same data in terms of depth resolution.
> On Jan 24, 11:47 am, Joanie<joan.l...@gmail.com> wrote: >> In PrimeSense's website (http://www.primesense.com/?p=514), the specs >> of a PS1080 SoC are:
>> Field of View (Horizontal, Vertical, Diagonal) = 58 H, 45 >> V, 70 D >> Spatial x/y resolution (@ 2m distance from sensor) = 3mm >> Depth z resolution (@ 2m distance from sensor) = 1cm >> Operation range = >> 0.8m - 3.5m
>> Questions:
>> 1) Do those above specs hold for the Kinect as well? (Since Kinect is >> also using PS1080)
Yes.
>> 2) The stipulated X/Y resolution is 3mm and Z resolution 1cm. But from >> the other posts in this forum, it seems like the accuracy is to 1mm >> for all X/Y/Z. OpenNI programs also output values to the mm. Does >> anyone know why the numbers are different?
The numbers usually come from lab tests on standard surfaces with a certain illumination, etc They will differ from surface type to surface type (i.e. some surfaces absorb IR more than others). There's many scientific papers on the subject (see http://polito.academia.edu/dariopiatti/Papers/172492/SR-4000_TOF_CAME... for example), and sometimes companies do put out datasheets with the actual details of the tests.
>> 3) Is the accuracy up to 3mm @2m for X/Y (similarly up to 1cm @2m for >> Z), and the accuracy decreases when you move further away from the >> sensor?
>> 4) The operation range is stipulated at 0.8m - 3.5m, but from another >> post, it says the Kinect output depth values at 30cm-10m. Why is there >> a difference?
The Kinect is using the same camera, same chipset, and (almost) the same firmware, so the numbers are not different. I am not sure who made statements about the Kinect producing anything useful at 10m. :) Sure, in theory it might be possible to get a depth value there, but it wouldn't be useful. Useful values are between 0.8 to 3-3.5m, as many have stated already.
>> It would be good if someone can give a more technical explanation on >> this please. Thank you!
1) Would you be able to explain why the XYZ values after 3.5m are not
useful? Is it because the XYZ data after 3.5m lose their accuracy?
Therefore, since @2m, the accuracy is 3mm (XY), 1cm(Z), the accuracy
after 3.5m is even lower than those numbers.
I guess the sweet spot is from 0.8m - 3.5m.
2) But OpenNI still outputs the XYZ values in millimetres. Therefore,
the numbers are actually somewhat interpolated.
Just FYI:
I am interesting in exploring it for longer range usage, and also need
precison. Hence the accuracy of these numbers are important to me.
On Jan 24, 3:16 pm, Radu Bogdan Rusu <r...@willowgarage.com> wrote:
> > The numbers for the kinect are more like 20mm X/Y resolution.
> > Microsoft significantly downscaled the resolution of the device to
> > make it cheaper.
> > The reference implementation would have a much better resolution.
> The Microsoft Kinect is basically the same sensor as the PrimeSense PS1080 using the same cameras and same chipset, with
> a "crippled" firmware (certain features have been taken out), so the two sensors produce exactly the same data in terms
> of depth resolution.
> > On Jan 24, 11:47 am, Joanie<joan.l...@gmail.com> wrote:
> >> In PrimeSense's website (http://www.primesense.com/?p=514), the specs
> >> of a PS1080 SoC are:
> >> Field of View (Horizontal, Vertical, Diagonal) = 58 H, 45
> >> V, 70 D
> >> Spatial x/y resolution (@ 2m distance from sensor) = 3mm
> >> Depth z resolution (@ 2m distance from sensor) = 1cm
> >> Operation range =
> >> 0.8m - 3.5m
> >> Questions:
> >> 1) Do those above specs hold for the Kinect as well? (Since Kinect is
> >> also using PS1080)
> Yes.
> >> 2) The stipulated X/Y resolution is 3mm and Z resolution 1cm. But from
> >> the other posts in this forum, it seems like the accuracy is to 1mm
> >> for all X/Y/Z. OpenNI programs also output values to the mm. Does
> >> anyone know why the numbers are different?
> The numbers usually come from lab tests on standard surfaces with a certain illumination, etc They will differ from
> surface type to surface type (i.e. some surfaces absorb IR more than others). There's many scientific papers on the
> subject (seehttp://polito.academia.edu/dariopiatti/Papers/172492/SR-4000_TOF_CAME...
> for example), and sometimes companies do put out datasheets with the actual details of the tests.
> >> 3) Is the accuracy up to 3mm @2m for X/Y (similarly up to 1cm @2m for
> >> Z), and the accuracy decreases when you move further away from the
> >> sensor?
> >> 4) The operation range is stipulated at 0.8m - 3.5m, but from another
> >> post, it says the Kinect output depth values at 30cm-10m. Why is there
> >> a difference?
> The Kinect is using the same camera, same chipset, and (almost) the same firmware, so the numbers are not different. I
> am not sure who made statements about the Kinect producing anything useful at 10m. :) Sure, in theory it might be
> possible to get a depth value there, but it wouldn't be useful. Useful values are between 0.8 to 3-3.5m, as many have
> stated already.
> >> It would be good if someone can give a more technical explanation on
> >> this please. Thank you!
OpenNI API supports values from 0 to ~32 meters, in millimeters. The PrimeSense PS1080 produces values between 30 cm and 10 m. No interpolation takes place. This means that there are values in this range that are not returned. For example (this is only an example, don't takes those numbers as facts) the depth values 2000 (2m) and 2010 (2.01m) can be returned, but all the values in between (2001-2009) will never be returned.
-----Original Message----- From: openni-dev@googlegroups.com [mailto:openni-dev@googlegroups.com] On Behalf Of Joanie Sent: Monday, January 24, 2011 19:58 To: OpenNI Subject: [OpenNI-dev] Re: Specification of PS1080 SoC
1) Would you be able to explain why the XYZ values after 3.5m are not useful? Is it because the XYZ data after 3.5m lose their accuracy? Therefore, since @2m, the accuracy is 3mm (XY), 1cm(Z), the accuracy after 3.5m is even lower than those numbers.
I guess the sweet spot is from 0.8m - 3.5m.
2) But OpenNI still outputs the XYZ values in millimetres. Therefore, the numbers are actually somewhat interpolated.
Just FYI: I am interesting in exploring it for longer range usage, and also need precison. Hence the accuracy of these numbers are important to me.
On Jan 24, 3:16 pm, Radu Bogdan Rusu <r...@willowgarage.com> wrote: > On 01/23/2011 08:48 PM, Dougx wrote:
> > The numbers for the kinect are more like 20mm X/Y resolution. > > Microsoft significantly downscaled the resolution of the device to > > make it cheaper.
> > The reference implementation would have a much better resolution.
> The Microsoft Kinect is basically the same sensor as the PrimeSense PS1080 using the same cameras and same chipset, with > a "crippled" firmware (certain features have been taken out), so the two sensors produce exactly the same data in terms > of depth resolution.
> > On Jan 24, 11:47 am, Joanie<joan.l...@gmail.com> wrote: > >> In PrimeSense's website (http://www.primesense.com/?p=514), the specs > >> of a PS1080 SoC are:
> >> Field of View (Horizontal, Vertical, Diagonal) = 58 H, 45 > >> V, 70 D > >> Spatial x/y resolution (@ 2m distance from sensor) = 3mm > >> Depth z resolution (@ 2m distance from sensor) = 1cm > >> Operation range = > >> 0.8m - 3.5m
> >> Questions:
> >> 1) Do those above specs hold for the Kinect as well? (Since Kinect is > >> also using PS1080)
> Yes.
> >> 2) The stipulated X/Y resolution is 3mm and Z resolution 1cm. But from > >> the other posts in this forum, it seems like the accuracy is to 1mm > >> for all X/Y/Z. OpenNI programs also output values to the mm. Does > >> anyone know why the numbers are different?
> The numbers usually come from lab tests on standard surfaces with a certain illumination, etc They will differ from > surface type to surface type (i.e. some surfaces absorb IR more than others). There's many scientific papers on the > subject (seehttp://polito.academia.edu/dariopiatti/Papers/172492/SR-4000_TOF_CAME... > for example), and sometimes companies do put out datasheets with the actual details of the tests.
> >> 3) Is the accuracy up to 3mm @2m for X/Y (similarly up to 1cm @2m for > >> Z), and the accuracy decreases when you move further away from the > >> sensor?
> >> 4) The operation range is stipulated at 0.8m - 3.5m, but from another > >> post, it says the Kinect output depth values at 30cm-10m. Why is there > >> a difference?
> The Kinect is using the same camera, same chipset, and (almost) the same firmware, so the numbers are not different. I > am not sure who made statements about the Kinect producing anything useful at 10m. :) Sure, in theory it might be > possible to get a depth value there, but it wouldn't be useful. Useful values are between 0.8 to 3-3.5m, as many have > stated already.
> >> It would be good if someone can give a more technical explanation on > >> this please. Thank you!
-- You received this message because you are subscribed to the Google Groups "OpenNI" group. To post to this group, send email to openni-dev@googlegroups.com. To unsubscribe from this group, send email to openni-dev+unsubscribe@googlegroups.com. For more options, visit this group at http://groups.google.com/group/openni-dev?hl=en.
