D435/D415 fabulous... but

[ROhle]

For Intel to produce the D400 series at this time seems like a miracle.

I have been working on a stereo analysis algorithm for over 30 years, but have never had anything close to the D415 and D435 available to me. I suspect that there are many others in a similar circumstance.

I have also never programmed in C++ or Python. So, I'm a complete newbie in this environment. And I am an amateur programmer. So, when you say "hacker" that overstates my credentials:)

Except for installing everything, the learning curve that the RealSense group has prepared for newbies is fabulous. It all works... mostly, except for issues that are normal for a development of this magnitude. The documentation is great. It is a hacker's dream.

# I have seen requests for scanning of various parts. In general, as is(out of the box), I'm afraid there is going to be a little disappointment. You can only go so far with the full logic and after that... hope for a hacker or two to come along.

Unfortunately, right now, the best routes to higher accuracy seem to be closed.

# We need full streams... all of them except depth. To get more accuracy, we need all of the raw data at maximum resolution... except depth. By moving to 10cm, I get great hands and faces, but my data is all binned and truncated... In Python2.7, I am stuck analyzing 640x480ir, when the cameras are capable of 1080p. I know I can get more from the Viewer, but I can't get all of it.

The Viewer used to allow color acquisition on Ir1... now it doesn't. Wrong direction!!! both Ir1 and Ir2 should allow full resolution color and the Color camera should drop the IR filter so we have 3 full channels to analyze. FPS is not critical for most scanning applications... but resolution is.

To change the Viewer interface you could simply segregate Depth and Stereo. When a particular Stereo resolution isn't supported by the Depth... you block access to Depth, just as is done now for other incompatible settings... but you allow the acquistions for other purposes.

#

One way of dealing with specular effects requires more than two cameras... We have three!!! But we can't get at them in a complete way.

I'm not talking about real time at this point... just accuracy and resolution... that's what many potential customers value most.

Thanks again

Rich

While I'm at it... I can understand limiting the search range to around 70 pixels... speeds things up and works great. What I don't get is building hardware that doesn't allow that 70 pixel search range to be used at any arbitrary distance. Same search time, better accuracy at near than we have now. Not trying to be fussy... I don't understand it and I would like to.

[MartyG]

Thanks for your great feedback, it's really appreciated that you took the trouble to provide that much detail.

Some people using Python with cameras other than RealSense who wanted to overcome the 640x480 default achieved it with OpenCV, using an OpenCV instruction called cap.set. For example:

cap = cv2.VideoCapture(0)

cap.set(3, 2560)

cap.set(4, 720)

More details can be found here:

https://docs.opencv.org/3.0-beta/doc/py_tutorials/py_gui/py_video_display/py_video_display.html https://docs.opencv.org/3.0-beta/doc/py_tutorials/py_gui/py_video_display/py_video_display.html

[ROhle]

MartG

You are the man! I'll take a look tomorrow. Dead tired.

Thanks

Rich