Advice on Choosing the Right ZED BOX for 4× ZED X Cameras at 30 FPS Recording

Hardware
, ,
1

Hello Stereolabs,

We recently purchased 4× ZED X cameras together with a ZED BOX Orin Nano. While the setup works, we have encountered a major limitation: during recording, the FPS drops significantly, and we are not able to achieve desired results.

Our goal is to record from all four ZED X cameras simultaneously, ideally maintaining close to 30 FPS per camera.

Before we consider upgrading, we would like to get some advice from Stereolabs :

  1. Which ZED BOX model is recommended for handling 4× ZED X cameras at maximum FPS (close to 30 FPS)?
  2. Perhaps you could recommend a particular build?
  3. I saw information in the Depth Modes section about Neural Depth Computational Performances on embedded devices . But if we rely on the code example, these results are not based on the svo record, correct?
  4. Is it possible to use DEPTH_MODE::NEURAL_LIGHT during recording and obtain point clouds based on NEURAL_PLUS mode during post-processing?

Thank you !

2

Hi @tiedvards

What model of the ZED Box Orin are you using? Nano or NX? RAM amount?

Do you want to perform this task while performing Depth extraction or not?

Only the Jetson AGX Orin model allows the extraction of depth information at 30 FPS with 4 cameras. We do not have ZED Box models powered by the AGX Orin module.

What do you mean?

You do not need to enable depth processing if not required while recording.
The SVO files do not include depth information, but only the RAW unrectified camera stream with sensor samples in the metadata.

Yes, as I said before, the SVO does not store depth information

3

We are using Jetson™ Orin Nano 8GB with 4x ZED X cameras.
I understand that there is (at least) a problem that this box lacks video encoding.

But let’s say I want to use a ZED Box Orin NX 8GB or Orin NX 16GB. Do you have any information on how many fps you could get when recording video with 4x ZED X cameras?

The main goal is to obtain point clouds from the recordings in order to subsequently assemble a model.
Based on your answer, as far as I understand, I don’t need DEPTH_MODE when recording, I can work with RAW recording and get point clouds in post-processing. Correct ?

Thank you for the answers !

4

Yes, this is correct.