Bookmark and Share

Hardware Topics

Path from Matlab/CameraDevBoard to Embedded Production System
2 replies [Last post]
lightyear66
Offline
Last seen: 4 years 46 weeks ago
Level 1: Prestidigitator
Joined: 2014-08-07
Points: 1

I have developed a simple prototype demonstration for an industrial machine vision application using Matlab on a PC and an Aptina Imager Dev Kit (MT9V024).  I'd like to take this concept to production in a small rugged form factor.  I need to deploy the system in a mobile field application where a large PC would maybe not survive and certainly not be very portable.  I could have a custom board designed and made with the imager and and maybe a Xilinx Zynq SoC and all the other supporting electronics to interface with the system.  This might provide the best performing solution at the lowest BOM cost.  But since my annual volumes will not be high (probably <100 per year), the development costs seem potentially prohibitive.  So, maybe I can buy an off the shelf camera wth USB out like The Imaging Source DMK 22AUC03 ( http://www.theimagesource.com/en_US/products/cameas/usb-ccd-mono/ ) and connect it to a single board computer (EPIC or EBX?) or similar stand-alone platform like ODROID-XU3.  Then I have to get my algorithm ported onto that system and hope that it can process the images in real time (>15fps).  My BOM cost may be ~3X higher, but the total cost per camera will probably be less. Has anyone done anything like this before that might offer some general or specific advice?  Maybe there is another path that has a good chance for success without huge custom development cost?  My background is in sensors and algorithms, especially in a dvlopment environment, with less experience with detailed hardware and final production implementation.  I likely need to find an experienced development partner for either approach to get a system working quickly with accepable risk.  Any recomendations for potential help that will not cost $100k+?

April
Offline
Last seen: 2 years 26 weeks ago
Level 1: Prestidigitator
Joined: 2016-11-16
Points: 5

connect it to a single board computer https://www.graperain.com/ARM-Single-Board-Computer/ (EPIC or EBX?) or similar stand-alone platform like Graperain. They way we would approach a project like this would be to first understand your algorithms, data types (e.g., do you need floating-point?) and data flows (e.g., do you have to process a full frame at a time?) - along with the degree of flexibility you have (if any) to adjust these things to enable a more efficient implementation.  From this we'd develop an estimate of the processing performance, memory bandwidth, and I/O capabilities required for an embedded implementation of your system.  This would then enable us to identify appropriate processors and platforms for a small, rugged implementation. You can come here for expert help.

Jeff Bier
Jeff Bier's picture
Offline
Last seen: 3 days 7 hours ago
EditorLevel 4: Thaumaturgist
Joined: 2011-05-29
Points: 93

Hi Todd,

My company, BDTI, specializes in finding the right processing platforms for embedded applications, and in porting and optimizing application software from PCs to embedded platforms.

They way we would approach a project like this would be to first understand your algorithms, data types (e.g., do you need floating-point?) and data flows (e.g., do you have to process a full frame at a time?) - along with the degree of flexibility you have (if any) to adjust these things to enable a more efficient implementation.  From this we'd develop an estimate of the processing performance, memory bandwidth, and I/O capabilities required for an embedded implementation of your system.  This would then enable us to identify appropriate processors and platforms for a small, rugged implementation.

Considering that your Aptina sensor is WVGA resolution and you're talking about 15 fps, and assuming that your algorithms are not extremely complex, I'm optimistic that there is an off-the-shelf platform that will meet your needs and will provide a reasonably easy programming environment.

I'd be happy to chat with you via phone to learn more about your requirements and give you some pointers to relevant resources.  And of course, if you're interested in engaging BDTI's services, I can provide you with more info on that as well.  Feel free to email me at bier[removethis]@BDTI.com.