Advice: ZED Mini Considerations for 3D object reconstruction and tracking at 30 cm working distance

[LolaB]

Hi,

I am currently evaluating ZED depth cameras for a system that requires 3D object reconstruction and online tracking

At this stage, I am considering the ZED Mini, and I would appreciate your guidance to better understand its suitability and possible alternatives within the ZED family.

My use case can be described at a high level as follows:

• Working distance of approximately 30 cm

• Object is curved and has low natural texture

• The solution needs to be eye-safe

• The camera is intended to be integrated into a larger system, with constraints on size, weight, and overall compactness

Based on this, I would be grateful if you could help answer the following questions by email:

• What depth and reconstruction accuracy can be expected from the ZED Mini at a ~30 cm working distance?

• Are there recommended approaches to improve reconstruction and tracking accuracy for curved, low-texture objects at this distance?

• Does adding artificial texture (e.g., patterns, speckles, or markers) typically improve performance in such scenarios?

• Given these constraints, is the ZED Mini the most suitable option, or are there other ZED models you would recommend for this type of application?

Thank you in advance for your support.

[Myzhar]

Hi @LolaB
Welcome to the Stereolabs community

@LolaB

What depth and reconstruction accuracy can be expected from the ZED Mini at a ~30 cm working distance?

Normally depth accuracy in optimal condition at 30 cm distance is 1%, so +/- 0.3 cm.
Please note that 30 cm is close to the minimum range value of the ZED Mini and image focus could not be optimal.

@LolaB

Does adding artificial texture (e.g., patterns, speckles, or markers) typically improve performance in such scenarios?

Yes, we use pure stereo vision with no Lidar or IR LED projectors, so texturization can add additional visual information that can help the surface reconstruction.

@LolaB

Given these constraints, is the ZED Mini the most suitable option, or are there other ZED models you would recommend for this type of application?

You could consider the ZED Mini which has a shorted baseline.

[LolaB]

Hi,

Thanks for your answer @Myzhar !

A couple of follow-up questions:

@Myzhar

Normally depth accuracy in optimal condition at 30 cm distance is 1%, so +/- 0.3 cm.
Please note that 30 cm is close to the minimum range value of the ZED Mini and image focus could not be optimal.

According to the ZED mini datasheet:

• Ideal Range 0.1m to 9m (0.3ft to 13.1ft)
• Depth Accuracy < 1.0% at 2m (6.6ft), < 1.8% at 4m (13.1ft)

Is the “~1% depth accuracy at 30 cm” figure based on actual measurements at 30 cm, or is it inferred/extrapolated from the 2 m spec? If measured, could you share the test conditions (lighting/texture, resolution/depth mode, calibration, etc.)?

@Myzhar

Yes, we use pure stereo vision with no Lidar or IR LED projectors, so texturization can add additional visual information that can help the surface reconstruction.

Could you advise what kinds of texturization work best in practice for improving surface reconstruction with ZED mini (e.g., random speckle pattern, high-contrast dots, checkerboard-like patterns), and any guidance on feature size/density at ~30 cm working distance?

[Myzhar]

@LolaB

Is the “~1% depth accuracy at 30 cm” figure based on actual measurements at 30 cm, or is it inferred/extrapolated from the 2 m spec?

We tested the camera in similar conditions.

@LolaB

If measured, could you share the test conditions (lighting/texture, resolution/depth mode, calibration, etc.)?

This information is not public.

@LolaB

Could you advise what kinds of texturization work best in practice for improving surface reconstruction with ZED mini (e.g., random speckle pattern, high-contrast dots, checkerboard-like patterns), and any guidance on feature size/density at ~30 cm working distance?

Normally, with classical stereo vision, random speckle patterns are the best condition. Repetitive patterns can create problems during the matching step.