How computer vision is helping logistics operators identify dangerous encounters before they become accidents, and why technology alone cannot eliminate the risk.
If you have worked on a distribution centre or a warehouse, you would notice how a forklift and a pedestrian rarely collide because a facility lacks rules. Most facilities already have marked walkways, mirrors at blind corners, horns, and trained operators.
Yet, pedestrians struck by forklifts are still the single largest cause of forklift-related fatalities, accounting for around 36% of the roughly 85 deaths OSHA estimates occur annually in the US, ahead of tip-overs and every other cause combined.
The controls exist.
What consistently fails is timing. A pedestrian stepping around a stack of pallets into a forklift’s path, or a reversing vehicle whose driver can’t see what a mirror doesn’t fully cover, in a conflict that opens and closes within a few seconds.
That’s the specific window AI-powered monitoring is being built to close, and after looking at how it’s actually being deployed, I think it’s worth walking through both what it changes and where it still comes up short.
Why the Problem Resists Fixed Controls
Vehicle-pedestrian risk concentrates in a handful of predictable spots, like loading docks, blind corners around racking, narrow aisles shared by forklifts and foot traffic, but the actual danger in any given crossing depends on variables that change second to second. For example, how fast the vehicle is moving, how much warning either party has, whether the pedestrian is in the driver’s direct sightline.
A safety officer doing scheduled walkthroughs isn’t present for the moment that matters. CCTV footage is useful afterward, but it doesn’t prevent anything on its own unless someone happens to be watching the right feed at the right second.
Teaching Cameras to Score the Interaction, Not Just Record It
Most large facilities already have extensive camera coverage, so the gap was never visibility, it was interpretation. A camera records a forklift passing two feet from someone the same way it records one passing ten feet away; an AI-based Vehicle–Pedestrian system built for this specifically tracks vehicle speed, pedestrian trajectory, and closing distance, then flags the interaction only when those combine into genuine risk.
A slow reverse near a worker gets treated differently than a forklift approaching a blind corner at speed. Some deployments add LiDAR specifically to hold up in low-light conditions, where camera-only distance estimates become less reliable but the underlying safety question- how close is that vehicle to that worker, right now? doesn’t go away.
What Real-World Deployments Reveal
The shift towards AI-supported monitoring is already showing up in how seriously manufacturers are treating it as infrastructure rather than experiment. Deloitte’s 2025 Smart Manufacturing and Operations Survey found that 28% of surveyed executives named vision systems among their top two investment priorities for the next 24 months, placing it behind only factory automation hardware and active sensors.
One example I came across was a Dubai-based energy equipment manufacturer employing more than 8,000 people. During the consolidation of several production units into a new facility, changing floor layouts created additional challenges for forklift operators and pedestrians who were still adjusting to unfamiliar routes.
The company introduced AI-based forklift monitoring to track forklift movements and pedestrian proximity, with alerts intended to help operators and supervisors respond to potential collisions. The manufacturer reported a 65% improvement in forklift safety following implementation.
For logistics operators, the potential benefits extend beyond faster hazard detection. Repeated alerts at a loading bay may reveal a poorly positioned pedestrian crossing. Concentrated near-misses during shift changes may expose traffic congestion or inadequate separation between workers and vehicles patterns a facility would otherwise have no reliable way of seeing.
Pedestrian detection in warehouse / Copyright: viAct
Where Computer Vision Still Falls Short
Computer vision cannot make up for poor warehouse layouts, inadequate pedestrian separation or gaps in operator training. Its accuracy also depends on conditions on the ground. Dust, glare, obstructed camera views and changing storage arrangements can affect detection, while excessive alerts risk becoming background noise for workers who encounter them throughout a shift.
There is also a difference between identifying a risk and doing something about it. AI may show that near-misses are occurring repeatedly at a particular loading bay, but it cannot determine every practical change needed to make that area safer. That still requires safety teams to review the findings, understand the circumstances and make decisions about traffic routes or working procedures. Unless these insights become part of daily safety reviews and operational planning, the technology risks becoming another monitoring system that collects information without changing much.
Privacy is another consideration, particularly when cameras continuously monitor areas where employees work. Measures such as face-blurring, anonymisation and compliance with applicable data protection requirements, including GDPR where relevant, can help address these concerns.
The value of computer vision ultimately depends on how well it is integrated into existing safety practices.
Where This Is Heading
The next development in computer vision is less about improving detection and more about what happens after an alert. Rather than a stream of individually scored warnings landing on a supervisor’s dashboard, some systems are starting to correlate recurring near-collisions against shift patterns and retrieve the relevant footage automatically, turning scattered alerts into an investigation-ready pattern instead of raw data someone has to piece together by hand.
For logistics operators, the next step is to make this analysis part of routine safety management.
About the author
Gary Ng is the Co-Founder and CEO of viAct, a technology company specialising in AI-powered workplace safety and operational monitoring. His work focuses on applying computer vision and AI to address safety challenges across manufacturing, logistics, warehouses and other high-risk industries. He is particularly interested in how emerging technologies can help organisations make better operational decisions and improve workplace safety.
