Machine Vision

cancel
Showing results for 
Search instead for 
Did you mean: 

Is this possible with GigE + PXIe hardware?

Hello,

 

A customer is asking for the ability to acquire 1900 x 1200 images (8-bit Bayer) at 30 fps. I've calculated this as

 

    1900 * 1200 pixels/frame * 8 bits/pixel * 30 frames/seconds = 547.2 Mbps (excluding TCP headers)

 

The PXIe-8234 can handle 1 Gbps per port, so this load should be ok in general, yes?

 

Now, the customer also asked for 6 cameras simultaneously, and having the images compressed to JPG and saved to disk (yes, 6 * 30 JPG images per second).

 

I've done this for 1280 x 1024 pixel images from 4 cameras simultaneously (2x PXIe-8234) at 30 fps -- a PXIe-8820 could just handle it (the CPU load was high), but the PXIe-8135 handled this comfortably. I'm not sure about 1900 x 1200 with 6 cameras though -- do you think this is possible? What are the likely bottlenecks in the system?

 

Thanks!

Certified LabVIEW Developer
0 Kudos
Message 1 of 4
(4,106 Views)

You need to convert the Bayer image to a color image somewhere in the process before you save it as a JPEG image.  The JPEG image compression destroys the details in the Bayer image, so it can't be converted later.

 

If you convert it on the camera, your image size will be 3 or 4 times larger.  This will result in a larger bandwidth requirement and looks like it would exceed your bandwidth.

 

If you convert it after transferring it to the PC, you are going to have a higher CPU load to handle the Bayer conversion.  The JPEG compression will also be a significant CPU load.

 

I don't know if it can be done.  I would doubt it is possible to do that much conversion, compression, and saving at that frame rate.

 

Bruce

Bruce Ammons
Ammons Engineering
Message 2 of 4
(4,066 Views)

Hi JKSH,

 

I may have an opportunity to plug 6x monochrom GigE cameras into either a PXI chassis with an 8135 or 8880 running Windows in the next few weeks. No guarantees yet though, but if I do, I can run a quick benchmark to lend some perspective on the CPU utilization which acquiring from all 6 cameras. Feel free to PM me if you need some more details.

 

Regards,

Joseph

Message 3 of 4
(4,059 Views)

Thanks for your responses, Bruce and Joseph.

 

To clarify, the customer specified 8-bit bayer simply because it is the cheapest coloured format for GigE streaming.

 

From the smaller system (4 cameras, 1280 x 1024 images), the bottleneck was indeed in the CPU operations. It wasn't disk I/O, because we got the same performance on both the standard PXIe-8820 hard drive (5400 rpm hard drive) and a custom solid-state drive. I'm wondering if the larger system would tilt the scales towards disk I/O being the bottleneck (I'll try to run some experiments and see)

 

Joseph, thank you for your offer. If you do get the chance to run some tests, I'd be most grateful if you could share your findings.

Certified LabVIEW Developer
0 Kudos
Message 4 of 4
(4,039 Views)