Maximum pixel disparity for depth measurement

Hardware
1

I’m considering the ZED X cameras for a robotics project and I am comparing them to other options as well as verifying depth performance, etc.

There may not be a simple answer given the system’s various depth modes. But is it possible to provide the maximum pixel disparity that the ZED X can handle (usually something like 128 or 256 pixels)? I can’t find this spec for the ZED camera systems.

Perhaps a simpler question. Will the ZED X camera measure depth at the minimum working distance of the spec sheet (0.3m for the ZED X and 0.1m for the ZED X Mini) in all resolution modes (including max resolution)?

Thanks!
Mike

2

Hi @michaelg
Welcome to the Stereolabs community.

The maximum disparity must be calculated based on various parameters using a specific formula:
max_disparity = focal_px * baseline_m / depth_maximum_distance_m

3

Thanks Walter! May I ask two additional questions for clarification?

In the provided formula, do you mean to have depth_minimum_distance_m (not maximum)? In my understanding of stereo vision, the disparity of an object will increase as it moves closer to the camera. Then, the maximum disparity that the stereo matching algorithm can use to estimate depth dictates the minimum working distance. On the datasheet the minimum working distance is given to be 0.3m (ZED X) or 0.1m (ZED X Mini).

Ultimately I want to know whether the minimum working distance on the spec sheet can be achieved at full camera resolution (1920x1080).

After more reading, I also noted the ‘minimum working distance’ setting in the software (Depth Settings - Stereolabs). The docs say that lowering this value increases processing requirements. I assume this is because the maximum disparity increases to process the larger range associated with a smaller working distance. A related second question then is how small can this value be set? Is the lower limit the minimum working distance given on the spec sheet, or can we go even lower (understanding there would be a drop in performance)?

So in summary two questions:

  1. What is the maximum stereo matching disparity in pixels? Or more simply, can the minimum working distance given in the data sheet be achieved at full camera resolution (1920x1200)?
  2. How small can the ‘minimum working distance’ software parameter be set, understanding that there will be a drop in performance? Is the lower limit the minimum working distance given on the data sheet, or can we go even lower if we accept a higher compute / memory load and/or lower FPS?

Again my apologies if I have not explained well or have misunderstood your reply. I’m new to stereo vision.

Thanks,
Mike

4

Yes, sorry for the typo. The correct formula is
max_disparity = focal_px * baseline_m / depth_minimum_distance_m

This is no longer valid, can you point me to the documentation that still reports this old information so we can remove it?

The minimum depth values for each camera are reported in the relative datasheets and depend on the mounted optics and the baseline.

5

@Myzhar thanks! Link to the minimum working distance setting in the SDK is in my quoted text above if your team wants to update.

Can I take this to mean that I can get stereo depth measurements at the minimum working distance while using full resolution images?

Thanks!
Mike

6

This is correct :+1: