Bookmark and Share

Hardware Topics

Smart Camera Design for Industrial machine vision
3 replies [Last post]
dpat123
Offline
Last seen: 11 weeks 17 hours ago
Level 2: Evoker
Joined: 2012-12-29
Points: 12

Embedded vision alliance is doing great work . I think community has to come up with a universal Smart Camera design .

BeagleBoard is once such plateform but this is not exclusively for Machine vision .

 

Compact design with Powerfull 1Ghz processor based plateform which should  accomodate CCD,CMOS and Linse scan sensor .

If community comes up with this type of hardware plateform it will a great leap forward.

dpat123
Offline
Last seen: 11 weeks 17 hours ago
Level 2: Evoker
Joined: 2012-12-29
Points: 12

VRagic D3 intellegent camera platefrom with 1Ghz Davinci TI processor is very interest development . Which is most compact industrial modular platefrom .

http://www.vrmagic.com/imaging/videos/

 

High end Emdedded vision application could be developed using this platefrom . It also had DSP so using dsplink one can also run dedicated code on DSP and take advantage for the same .

 

Amit Shoham
Offline
Last seen: 3 weeks 4 days ago
Level 4: Thaumaturgist
Joined: 2011-05-28
Points: 60

I agree that a community-based platform for machine vision would be great.

In addition to the ability to directly connect CCD/CMOS/Line-scan sensors, it should support at least two sensors simultanesouly, with synchronization, for stereo vision applications.

Another key piece of the puzzle is that many vision applications need some acceleration outside of the ARM CPU, e.g. on a GPU or DSP.  And then the tools and software infrastructure becomes much more complex.  I think that this is the biggest hurdle to overcome in providing a vision-specific platform.  Without the acceleration, many applications are so limited that there's not much point in providing dedicated camera interfaces--connecting a USB camera to a Beagleboard provides as much resolution and bandwidth as the application can handle.

So ideally we'd want an application processor with a GPU and/or DSP, libraries such as OpenCV/FastCV that are specifically optimized for this application processor, and readily available tools.

I suspect that we're going to see a lot more options in the near future in terms of hardware, e.g. more things like the zed board for the Xilinx Zynq, but the software and tools support will lag behind for a while and getting optimized performance will be very challenging.

-Amit

stillwater
Offline
Last seen: 19 weeks 6 days ago
Level 1: Prestidigitator
Joined: 2013-02-04
Points: 9

In addition to the problem of supporting a universe full of sensors, the demands for non-trivial vision applications, makes the goal of creating an affordable community-based platform a stretch goal.

If I look at the success stories in community-based platforms, I see just one, Arduino, and that is mostly as a supersimple integrated MPU with A/D and D/A. The DIY UAV folks have build a nice software stack for a particular interesting use case of building low-cost rotor-based UAVs.

A platform that could take off is the Pebble watch that connects with your cellular infrastructure and creates a hardware platform for interesting health and personal activity tracking. 

In short, the community needs a well defined application that is of interest to a large swath of the embedded vision developers before you can define a reasonable community-based hardware platform to support that application.

For additional emphasis: the application needs to come first.