image recognition benchmark

[Alex2019]

Dear machine vision experts,

Have anybody worked with NI Compact Vision/Industrial Controllers? Like those listed in http://www.ni.com/sv-se/shop/select/compact-vision-system 

My question is about benchmark/performance. In particular, for my customer, I need to know how many frames/second (of the reasonable image size, not HD) I can process? For example, a simple line (bright and dark stripes) recognition problem. Have anybody quantified performance with CVS-1459RT or similar?

[muhammad.kamran.ayub]

Hey Alex,

This really depends on many things, including, camera interface type, hardware, where you want to do processing, etc.

We normally have 2 types of vision specialized systems:

1. CVS (normally take image on host (RT Processor)

2. FlexRIO/FAM based (acquire directly on FPGA), can offer significantly higher data rates and processing

You can get a good FPS with USB 3.0, thus acquiring images at host and processed at FPGA of CVS. A bottle neck here is USB 3.0 data rates and of course RT-FPGA bi-directional data transfer rates. (Just in case you do not have background, NI CVS systems have FPGA and RT Processor in one place). I like FPGA  for extensive processing as host might not be able to handle. The rest depends on how optimized your FPGA image processing code is. Here is starting point of FPGA based image processing.

If these bottle necks affect your application and you want more high data rates and direct FPGA acquisition and processing, FAM should be your choice.

Please let me know if you have any particular case/scenario to discuss.

Muhammad Kamran Ayub
Staff Technical Support Engineer | Platforms & Systems

Certified LabVIEW Architect (CLA) | Certified TestStand Developer (CTD)

National Instruments UK & Ireland

[Alex2019]

Thank you very much Muhammad for your valuables points!