Skip to content
TheChatArchiveThe Chat Archive
Geminigemini-2.5-pro· GoogleHugging Face LMArena Human Preference 140K

What measurement accuracy in mm does the Zed 2i have at various distances?

From a Chatbot Arena battleThe same person sent this prompt to two anonymous models at once. Compare it with Hunyuanhunyuan-turbos-20250416’s answer.
  1. User

    What measurement accuracy in mm does the Zed 2i have at various distances?

  2. Gemini Assistant

    Of course. This is an excellent and crucial question when working with any depth camera.

    The measurement accuracy of the ZED 2i is not a single, fixed number. It fundamentally depends on the distance to the object being measured. The further away an object is, the less accurate the depth measurement becomes.

    Here’s a breakdown of the ZED 2i's accuracy at various distances.

    The Rule of Thumb

    A good rule of thumb provided by Stereolabs for their cameras under optimal conditions is:

    The depth error is approximately 1% of the distance.

    This is a simplified but very useful starting point for estimations. However, the error actually increases with the square of the distance, so this 1% rule is more of an average and becomes less reliable at very long ranges.

    Accuracy Table at Various Distances

    Here is a table that translates this rule into practical numbers and adds some real-world context. The ZED 2i is designed to work from 0.2m to 20m.

    Distance to ObjectTheoretical Error (using <1% rule)Practical Error in Millimeters (mm)Use Case & Notes
    0.5 m (500 mm)< 0.5%< 2.5 mmExcellent: Ideal for object inspection, robot arm guidance, close-up metrology.
    1 m (1000 mm)< 1%< 10 mmVery Good: High-quality 3D scanning, obstacle avoidance for indoor robots.
    2 m (2000 mm)< 1%< 20 mmGood: People tracking, gesture recognition, room scanning.
    5 m (5000 mm)< 1%< 50 mmFair: General obstacle avoidance, large-scale environment mapping. Individual points are less precise.
    10 m (10,000 mm)~2%~200 mm (20 cm)Low: Suitable for detecting large objects (e.g., cars, walls) but not for precise measurements.
    15 m (15,000 mm)~3-4%~500 mm (50 cm)Very Low: Detection of very large obstacles. The depth map will be noisy and imprecise.
    20 m (20,000 mm)>5%> 1000 mm (1 m+)Detection Only: This is the maximum recommended range. Used for presence detection, not measurement.

    Why Does Accuracy Decrease with Distance?

    The ZED 2i is a passive stereo camera. It calculates depth by finding the same feature point in both the left and right images and measuring the difference in their pixel positions (this is called disparity).

    • Close Objects: Have a large disparity (the point appears in very different locations in the two images). A small error in finding the exact pixel location has a minimal impact on the final depth calculation.
    • Far Objects: Have a very small disparity. The same one-pixel error in measurement now leads to a massive error in the calculated depth.

    This relationship is described by the formula:

    Depth Error ∝ (Distance)² / (Focal Length × Baseline)

    This formula shows the two key takeaways:

    1. Quadratic Error: The error grows with the square of the distance. Doubling the distance quadruples the error.
    2. Baseline Matters: The ZED 2i has a wide 120 mm baseline (the distance between the two cameras), which is a major advantage for accuracy, especially at mid-to-long ranges, compared to cameras with smaller baselines.

    Key Factors That Affect Real-World Accuracy

    The numbers above assume ideal conditions. In practice, several factors will impact your results:

    1. Lighting: The camera needs good, consistent lighting to see features. Poor light increases image noise, which directly hurts accuracy. Glare or over-exposure will wash out features.
    2. Texture: The system needs texture to match points. It will fail on uniform surfaces like a blank white wall, a plain floor, or a pane of glass. The ZED SDK's NEURAL Depth Mode is significantly better at handling these challenging surfaces than older, traditional stereo algorithms.
    3. Camera Calibration: The ZED 2i comes factory-calibrated. For high-precision applications, you can re-calibrate it using the ZED Calibration tool to ensure optimal performance.
    4. ZED SDK Settings: The Stereolabs SDK offers different depth modes that trade performance for accuracy.
      • NEURAL: Best quality, especially for challenging surfaces.
      • ULTRA: High-quality depth map using classic stereo methods.
      • QUALITY / PERFORMANCE: Faster modes with lower accuracy, suitable when high frame rates are needed.
    5. Motion: Fast motion can cause motion blur, making it harder for the algorithm to match features accurately. The ZED 2i's global shutter helps minimize this compared to rolling shutter cameras.

    Summary

    For the ZED 2i:

    • Excellent Accuracy (< 2 cm): Within 3 meters.
    • Good/Usable Accuracy (2-10 cm): Between 3 and 7 meters.
    • Rapidly Degrading Accuracy: Beyond 7-10 meters, it's best used for detecting the presence of large objects rather than for precise measurements.

Responses were produced in different, uncontrolled contexts (system prompts, settings, prior turns). This is not a benchmark.