Warehouse Near-Miss Detection Guide
Why near misses matter
Most warehouse safety programs have more close calls than recorded injuries. That is good news for the people who walked away from the event, but it creates a measurement problem. If a forklift brakes hard near a pedestrian, if a pallet jack cuts through a walkway, or if a worker steps out from behind racking with no collision, the event often disappears. No one writes it down. No one reviews the camera. No one sees the pattern until the same aisle produces a reportable injury.
Near-miss detection fixes that by treating close calls as safety data. DHI uses the cameras already watching aisles, docks, and staging lanes to identify the moments where paths, speed, distance, and visibility create risk.
What a useful near-miss record includes
A useful record is not just a clip. It needs enough context to change behavior. The event should include the camera, aisle, timestamp, object types, direction of travel, and a reason the system flagged the moment. A clip without metadata becomes another video to watch. A clip with metadata becomes a trend.
Aisle and zone labels
Start by naming zones the same way supervisors name them. Dock 4, north cross-aisle, battery charging, outbound staging, and cold storage entrance are better labels than camera 17. The point is to make the record usable in a shift meeting.
Speed and path context
Forklift safety is not only about whether a vehicle and person appeared in the same frame. The useful signal is whether their paths were converging, whether a stop was sudden, whether a pedestrian was occluded, and whether the vehicle entered a walk path too quickly.
Where to start
Do not begin with every camera. Pick the zone with the most complaints, the most supervisor concern, or the clearest history of close calls. In most facilities that means a cross-aisle, dock door, staging lane, or blind corner near high-pile storage.
Camera angle review
A camera mounted for general security may not see enough of the floor to support near-miss detection. Before the pilot begins, walk the zone and confirm that the camera sees the travel path, pedestrian path, and conflict point. If the floor zone is cropped out, the model cannot infer the same risk a person on site understands.
Response ownership
Near-miss detection creates value only if someone owns the follow-up. Decide before launch whether the event goes to a safety manager, shift supervisor, training lead, or VMS operator. Then decide which events deserve immediate review and which belong in a weekly trend report.
What to measure during a pilot
A warehouse near-miss pilot should produce a small set of practical numbers. Count repeat hot spots by aisle, repeat times by shift, events by incident type, and how often staff agree the clips were worth reviewing. That last number matters because a system that overwhelms supervisors with weak clips will not survive.
Turning close calls into prevention
The goal is not to punish workers for being on camera. The goal is to identify where the environment is producing the same risky interaction again and again. If near misses concentrate around one rack end, one dock lane, or one shift change, that is a layout and process signal. The fix might be new signage, a local light, driver retraining, a one-way walking path, or a camera angle change.
DHI makes that review possible without asking a person to watch hours of footage. The cameras were already there. The missing layer was the ability to turn the few risky seconds into a record safety teams can use.