Using the R200 with a single board computer (Minnowboard Turbot)

[kfind1]

Hi there! I saw some previous forum topics on using the Realsense 3D cameras with small computers such as the Intel Compute Stick (https://software.intel.com/en-us/forums/realsense/topic/589696)  and some others asking about using Atom CPU in general (https://software.intel.com/en-us/forums/realsense/topic/600759).

I do not plan on using the advanced features of Scene Perception module which for some silly reason only works if you have Open CL 1.2 even if you wanted to use parts of the module which don't actually need/use that functionality. Therefore I am not worried about some of the other "requirements" of the current SDK. 

I plan on simply grabbing raw depth, calibrating using the camera's built in StreamCalibration and projection functionality, and using the PCL library to process the point cloud. 

From those previous forum topics I have found a newer version of the Minnowboard MAX called the Minnowboard Turbot. This is basically the same, but with a few small improvements and some certifications. It has a USB 3.0 host, and an Atom 1.46Gz dual core CPU (Bay Trail). It seems to tick all the boxes for hardware requirements/recommendations except the Cherry Trail issue - but samontab says he has got raw depth etc from an Atom CPU which may or may not be a Cherry Trail. The recommended spec for the current (2016 R1) SDK is an Intel Haswell architecture CPU, but my laptop is doing just fine running the SDK and applications on an old i5-2450M, so I'm not sure how exactly bad it will be not having a Cherry Trail.

note: for those worried about memory footprint (RAM usage) my applications so far haven't used more than 350MB during processing, so a single-board computer with 2GB RAM is more than enough.

Anyone who has proven they can run a simple (depth data preferrably) Realsense application on a Bay Trail Atom CPU would be awesome to hear from. 

Also, it should be mentioned to those interested that the new model of the Intel Compute Stick now has USB 3.0 and a quad core Atom X5 (meaning, it's Cherry Trail!) so it's now in line with the hardware requirements of the R200. 

In addition to any (new) thoughts on what i'm trying to do, I am interested to know when x86 (32 bit) and linux OS will be supported by the Realsense SDK, I think I read this would be on the next SDK release?

edit: One final question - what is the likelihood a powered USB3.0 hub for 3-4 R200s connected will work? The single-board-computer (SBC) connects to the hub with its USB 3.0 host, then multiple R200s connect to the hub and they are mounted in static locations around the work area, with overlapping fields. The SBC then can connect to all of them simultaneously probably using librealsense, and take depth images from each one in a round robin style to build up a single global point cloud. This is not an application where moving the R200 is possible that's why i'm asking about this option.

[samontab]

A few things:

Minnowboard Max or Turbot should in theory be able to run any realsense camera fine, although you never really know for sure until you try, specially because of power issues. This is even more critical when you plan to use more than one camera.

I've used F200 and R200 with an Intel Atom CPU Z3775@1.46GHz for grabbing raw depth data and color, and it works fine.

For using this camera in Linux, and other OSs, you should defitely look at the librealsense (official from Intel, just gives you raw capabilities):

https://github.com/IntelRealSense/librealsense

[kfind1]

Thanks samontab, indeed i've already obtained the latest build of librealsense from github. I may convert my current SDK-based program to using that before I get too far into my development.

I will get a Minnowboard Turbot and test it out. As for power, I think a powered USB hub will help with that issue - my own custom 5V 6A supply should be enough to power the hub plus the minnowboard and however many USB3.0 devices I want. The R200 only uses 300mA supposedly. 

Thanks for confirming it works on a Z3775, that is comforting. 

[samontab]

No worries.

It would be great if you could report here if the Minnowboard actually works fine or not when you get to test it.
 

[kfind1]

samontab wrote:

No worries.

It would be great if you could report here if the Minnowboard actually works fine or not when you get to test it.
 

So my Minnowboard Turbot with Windows 10 x64, can connect to an R200 and with DCM seemingly installing itself all I did was install the SDK R1 2016 (8.0.24.6528) and run the Sample Browser, so far I have successfully run the Camera Viewer and Projection samples, and the Camera Explorer tool. Great news! These are the only functions of the SDK I am interested in, but I could run some kind of blob or hand detection or segmentation samples too. 

I will be trying sooner or later Ubuntu 14.04.4 LTS with the required kernel upgrade to 4.4 and I will try to run some librealsense samples/example programs on the Minnowboard, but i'm not sure when i'll get around to doing that. 

[samontab]

Thanks for confirming that the R200 works fine on the Minnowboard Turbot!.

[DKryd]

kyran f. wrote:

Quote:

samontab wrote:

No worries.

It would be great if you could report here if the Minnowboard actually works fine or not when you get to test it.
 

So my Minnowboard Turbot with Windows 10 x64, can connect to an R200 and with DCM seemingly installing itself all I did was install the SDK R1 2016 (8.0.24.6528) and run the Sample Browser, so far I have successfully run the Camera Viewer and Projection samples, and the Camera Explorer tool. Great news! These are the only functions of the SDK I am interested in, but I could run some kind of blob or hand detection or segmentation samples too. 

I will be trying sooner or later Ubuntu 14.04.4 LTS with the required kernel upgrade to 4.4 and I will try to run some librealsense samples/example programs on the Minnowboard, but i'm not sure when i'll get around to doing that. 

kyran, did you use a desktop windows 10 x64 build and install via external usb cdrom drive or use a windows 10 x64 IoT build , a preview release from microsoft.  also was there any firmware you flashed prior to win10 install? thanks.

[kfind1]

Douglas, I used the Windows 10 x64 desktop build, first on a micro SD card using "windows to go" to install it to the SD card. Then, after finding that Windows 10 is extremely poor on an SD card (too much file read/writes, SD card gets overloaded and huge delays) I reinstalled windows 10 on a spare 128GB SSD and use SATA into the minnowboard, and the 5V fan header J2 to power the SSD. Windows 10 now runs extremely smooth, it's quite awesome.

I did not flash any bios or firmware - my minnowboard turbot has firmware v82 which seems good enough, even though v90 is available now. The BIOS and all default options worked fine.

I am avoiding the use of Windows 10 IoT because I have some heavy duty desktop-style applications and I doubt ( I don't actually know for sure ) that the Point Cloud Library (PCL1.7.2) and OpenCV 2.4.10 will run on it.

I am very excited for the upcoming Windows 10 insider builds which supposedly released March 30 (but I have yet to witness my laptop install the build..) that has the support for the "universal" windows application layer - allowing our desktop Windows 10 OS to run embedded drivers properly and we can get seamless (IoT style function calls) access to the GPIO, UART, and I2C driver subsystems on the minnowboard from user applications.

[DKryd]

kyan,

well today i did a windows 10 install. off a dvd drive  onto a turbot. using a lure with a mSata SSD. it surprised me how well the install went after reading a few forums/lists i was expecting problems. prior to the win10 install i tried an ubuntu install off an usb thumb drive,and got boot failure on all attempts. i will probably try a dvd install later. i'm gonna try my R200 later tonight. i'm thinking i might get a nice ultra small  "portable" hardware setup. not to sure about running everything off a battery pack. any ideas about battery?

[kfind1]

Hi Douglas (I see you posted on the mailing list for Minnowboard too :-)  )

You can certainly run everything off a battery pack, the only thing you'll need is a high efficiency buck converter to convert the battery voltage down to 5V with approx. 3A max should be good enough. Depending how much run-time you want, I would select a battery pack from those requirements.

With a buck converter you will want a ~2V dropout between 5V and the battery voltage, so go for a battery pack that always gives more than 7V throughout it's charge.  I suggest a 3 cell lithium-ion battery pack with ~2 Amp-hours of capacity. If we guestimate the Minnowboard power usage plus your mSATA drive to be 1A on the 5V input, and the R200 uses 300mA during operation, this is a total of 1.3A (6.5W @ 5V).

With perfect conversion from battery voltage to 5V, assuming 3 cell (11.1V nominal) Li-Ion battery at 3Ah capacity gives you 33.3 Watt-hours of usage, you can run a 6.5W system for just over 5 hours (5 hours 7 minutes). If you account for a poor conversion efficiency of a buck regulator, down-rate your run-time to account for losses, lets say 15% loss. This gives you 4 hours and 21 minutes runtime with honestly pretty high expected load - the minnowboard + mSata drive should use less than 1A so the 4 hours 21 minutes is a pretty conservative minimum runtime.

Here's an example of the Lithium Ion battery pack you might use: Link to hobbyking product page

Here's an example of a small buck converter which could work: Link to Pololu product page for 2.5A output,  or for 5A output