Visual AI in Manufacturing: Building Reliable Near-Miss Detection - August 12, 2026

Name: Visual AI in Manufacturing: Building Reliable Near-Miss Detection - August 12, 2026
Start: 2026-08-12
End: 2026-08-12

Aug 12, 2026

10:30 AM - 12:30 PM PST

Online. Register for the Zoom!

About this event

Manufacturing teams are under pressure to move safety programs from after-the-fact incident reporting to earlier detection of the conditions that lead to serious injuries. While many facilities already have the raw video footage needed for this shift, the harder problem is making that footage usable.

Near-misses are rare, fast, and highly context-dependent. A clip that looks unsafe in one plant may be normal work in another. This session focuses on the practical workflow behind reliable near-miss detection: how to define the events that matter, find them in large volumes of ordinary footage, review ambiguous cases, reduce false alarms, and evaluate model performance across plants, cameras, shifts, layouts, and operating conditions.

You'll leave with a clear framework for turning plant-floor video into a safety-critical dataset and feedback loop that can support model improvement, operational review, and trustworthy deployment.

Host

Topics include how to:

Define “near-miss” and “unsafe proximity” in a way that both safety leaders and computer vision teams can use
Understand why reliable near-miss detection is harder than generic person, vehicle, or PPE detection
Mine long-form plant video for rare candidate events without labeling everything frame by frame
Handle ambiguous safety labels, reviewer disagreement, and site-specific definitions of risk
Evaluate a safety model beyond aggregate accuracy, including missed hazards, false alerts, camera placement, lighting, occlusion, equipment type, and facility layout
Reduce alert fatigue by separating meaningful hazard precursors from normal plant activity
Ensure human review, automation, and agent-assisted workflows can work together without removing expert judgment from safety-critical decisions
Build a repeatable feedback loop from model failures back to better data, better labels, and more reliable deployment

Who should attend:

ML and Computer Vision Engineers building industrial video systems
EHS and safety leaders evaluating AI for near-miss detection and serious-injury reduction
Manufacturing operations, automation, and digital-transformation teams working with plant-floor video
Data and annotation teams responsible for turning safety definitions into usable training and evaluation data
Safety technology teams moving from pilot deployments to reliable production systems