VisionAery

    V4 · Visible Vapor Detection

    Visible Vapor Detection

    An early-warning layer for releases that form a plume your camera can see. V4 processes the image at the edge, keeping detection on the camera or on a server at your site.

    Compressor Station Vapor Release
    Compressor Station Vapor Release
    Runs onIntel

    Visible Vapor Performance Benchmark

    Spray and Visible Vapor — Combined Benchmark

    VisionAery V4 Liquid Leak Detection was trained on over 200,000 images.

    Accuracy was evaluated with 9,147 previously unseen frames, without per site training.

    Read the Benchmark Methodology
    98.1%
    Detection Rate
    3.7%
    False Positive Rate
    0.923
    F-score

    Spray and visible vapor were evaluated together; these are not separate measurements for either type. Detection and false positive rates are frame-level results, not alarm rates or performance guarantees. F-score is the pixel-level harmonic mean of precision and recall. Scenarios outside those tested are not covered by these figures.

    Detection Across Industries

    Visible Vapor Customer Use Cases

    Monitor for visible vapor day and night, using artificial lighting or the camera’s near-infrared (NIR) illumination after dark.

    Visible Vapor Detection Questions

    Can it detect invisible gas?

    No. A release must form a visible plume the camera can see. VisionAery does not identify gas species or measure gas concentration.

    Was visible vapor measured separately from spray?

    No. The published 98.1% frame-level detection rate, 3.7% frame-level false positive rate and 0.923 pixel-level F1/Dice cover spray and visible vapor combined. These model results are before the alarm engine, not separate vapor-only measurements or alarm-rate guarantees.

    Does this replace gas detectors?

    No. It is a camera-visible early-warning layer, not a replacement for required gas instrumentation, inspections or safety procedures.

    How was the model trained and evaluated?

    VisionAery V4 Liquid Leak Detection was trained on over 200,000 images. The held-out benchmark used 9,147 previously unseen frames without per-site training.