djtoshi182

http://www.youtube.com/watch?v=wS8wyIYn77w

I would like to make this look that generates depth of field in grayscale.

Also, can I set the distance range from the nearest to farthest from the depth?

I would like to make this look that generates depth of field in grayscale.
Could you guide me to make this patch?
I have jit.freenect

kinect-depth-of-field-control-in-grayscale

If you are windows, then use the jit.openni external. Easy to use, included demo patch, and the wiki there is full documentation.

Use the output of the depth outlet, do some interpolation to smooth the grey, and then cut out depth values further back than you want (e.g. depth values larger than 1000 mm).

You could set the min/max values using OpenNI's XML config file, however, I have not seen it add any significant improvement in depth resolution. The docs to set such values is in the OpenNI documentation that gets installed with the OpenNI SDK.

Thanks for the help. although I am using mac

I see 'Kinect RGB' features depth range setting and color mapping.

http://nicolas.burrus.name/index.php/Research/KinectRgbDemoV4?from=Research.KinectRgbDemo

I'd like to set the depth range from 30cm~at least 5m, also in grayscale.

plus, can I invert the grayscale output from kinect? (w>b to b>w)

Any other Kinect resolutions for Jitter?
I see 'Kinect RGB' features depth range setting and color mapping.

I'd like to set the depth range from 30cm~at least 5m, also in grayscale.
plus, can I invert the grayscale output from kinect? (w>b to b>w)


As I understand, Kinect has a hardware limitation of no closer than 500mm. And that the depth sensor can get a map of up to 10m. FYI, skeletons are not reliable out that far...only about 4m.

Depthmaps are 10bit of depth info/pixel. You can interpret it however you want. For example, you could send it a pwindow and you'll see a greyscale image. You could use jit.op to apply matrix operations to it to achieve the inverting you want. Other ways to do it also.

do you have any ideas regarding ways to remove (trim) points from a matrix based on value?

i had a hack with sorting the matrix (using jit.xray externals) and re-sizing it,

hey diablodale, somewhat related
do you have any ideas regarding ways to remove (trim) points from a matrix based on value?
i had a hack with sorting the matrix (using jit.xray externals) and re-sizing it,
but its too slow for big matrices.


The matrix itself is just memory locations w/ values in them. You could resize the matrix and throw away rows/columns/3rd/4th/etc dimensions. jit.findbounds could help you find value ranges in a larger matrix that you want and then use jit.submatrix to chop it.

However, are you asking if you could interpret the values in a matrix to create some kind of mask. For example...if you had a matrix of depth values and you wanted to "ignore" values more than 4m?

If so, then first choose what "ignore" means. Does it mean that the value of that matrix location is set to zero. does it mean that only the alpha is set to zero. AFter that, you can use matrix operators like jit.op or jit.expr to first test your pixel for "removal" and then set its value.

If you were to use my jit.openni object, it provides a depthmap matrix and also a separate usermap matrix. By using the latter, its each to mask out everything except the desired user.

Attached is a patcher that takes these two matrices: depthmap on left, usermap on right. The output is a matrix where only user the user with id=2

is visible; I set the alpha to zero on all other pixels.

Attached is a patcher that takes these two matrices: depthmap on left, usermap on right. The output is a matrix where only user the user with id=2
is visible; I set the alpha to zero on all other pixels.


exactly what I am trying to sort out right now: get a usermap, but restrict the depth to let's say 3meters, so the users can hide behind an imaginary line.

My first strategy would be to do that with the XML config file. I mentioned my trials in the other thread already. I have to admit that I think I failed because I thought the OpenNI pdf would give the same info as the *.chm. Silly me, in the *.chm all the developer info can be found.

It does not work yet, but I assume when it is in the documentation it MUST work, right?

I see code attempting this in the XML you provided. Did you get it to work?

(Sidenote: my version does not "cut off the heads". I will find out what makes the difference and will post.)

The second strategy would be to do as suggested here. With my knowledge I would work on the depthmap, use everything above or below the desired value and then merge it with the usermap. Definitely working, not a big deal using e.g. Vizzie Oper8r. But not really elegant, what you suggest above (jit.op @op && pass pass pass) looks easy to handle, BUT: not working at all, no matter how I try it. Any suggestions before I post the patch?

Hello Dale,
exactly what I am trying to sort out right now: get a usermap, but restrict the depth to let's say 3meters, so the users can hide behind an imaginary line. 

My first strategy would be to do that with the XML config file. I mentioned my trials in the other thread already. I have to admit that I think I failed because I thought the OpenNI pdf would give the same info as the *.chm. Silly me, in the *.chm all the developer info can be found.
It does not work yet, but I assume when it is in the documentation it MUST work, right?
I see code attempting this in the XML you provided. Did you get it to work?
(Sidenote: my version does not "cut off the heads". I will find out what makes the difference and will post.)

The second strategy would be to do as suggested here. With my knowledge I would work on the depthmap, use everything above or below the desired value and then merge it with the usermap. Definitely working, not a big deal using e.g. Vizzie Oper8r. But not really elegant, what you suggest above (jit.op @op && pass pass pass) looks easy to handle, BUT: not working at all, no matter how I try it. Any suggestions before I post the patch?


What about using something like
(jit.op @op 

here's the technique i use on the mac to get jit.freenect.grab to give me a normalized depth map between closest element and other point:

KINECT depth of field control in grayscale