Need advice to use 4 cameras D455 on a same computer

[Shooobaka]

Hello,

For the purpose of data acquisition, with my coworkers, we plan to use 4 Intel RealSense D455 cameras connected to the same computer to capture frames (both RGB and depth) with a resolution of 1280x720 and a framerate of 30 FPS.

For now, we did the test on a computer with a GTX 1080 Ti and 32 Gb of RAM. We can use 2 D455 (3 is possible but often causes freezes) and can go up to 640x480 with 30 FPS with no problem (1280x720 causes freezes). We would like to order a dedicated computer, but we are stuck trying to figure out what would be the best hardware options for a project like this. By any chance, do you have recommendations ?

Thanks!

[MartyG-RealSense]

Hi @Shooobaka The more RealSense cameras that are simultaneously active on the same computer, the more of the computer's resources that will be consumed. For four cameras, a computer with a processor equivalent to an Intel Core i7 or better is recommendable.

32 GB of RAM is a very good amount of memory.

If you are experiencing freezes at 1280x720 on your test computer then a good video card / GPU chip is likely to be a priority if your project is unable to use a lower resolution.

If you are going to be writing your own RealSense programs and using C++ language then using the RealSense SDK's GLSL feature may provide a performance benefit by automatically offloading some processing work from the processor to the graphics chip. GLSL is 'vendor neutral' and so should work with any graphics card / GPU chip, though performance improvement may not be noticable on low-end video hardware. #3654 describes the pros and cons of using GLSL and when to use it.

[Shooobaka]

Hi @MartyG-RealSense Thanks for your answer. We are using OpenNI2 with librealsense drivers, the RealSense SDK is already built with the parameter BUILD_GLSL_EXTENSIONS set to ON, is it enough to enable it?

[MartyG-RealSense]

A C++ script would not use GLSL unless the RS2:: instructions in the script are deliberately modified to insert GL:: after RS2:: - such as RS2::POINTCLOUD being changed to RS2::GL::POINTCLOUD

As you are using OpenNI2, using librealsense's in-built GLSL graphics acceleration support will therefore not be an option unfortunately if using an OpenNI2 script instead of a RealSense C++ script.

[Shooobaka]

Ow, too bad. Again thanks for your answers, it will definitely help us.