Video Safety
Live verdicts on streamed video
Four prebuilt alerts and four rules the customer wrote, each returning the reason and where in the frame it fired.
Video safety evaluates a chunk of streamed video rather than a single still. The same plain-English policy model applies: a policy is saved once, the platform compiles it, and each chunk comes back with a verdict, the reason the verdict was reached, and where in the frame the evidence was found.
Two kinds of rule run against the same stream. Prebuilt alerts are named events
the platform ships with, carrying a severity. Customer policies are rules the
enterprise writes in its own words, and they return blocked against the
policy that fired.
Prebuilt alerts
Named events, each returning a severity and a location in the frame.
Two individuals are seen engaging in a physical struggle involving striking and grappling movements.
center of the frame across the first four tiles
A thick, dark plume of smoke is filling the upper half of the frame and obscuring the background.
above the tree and building
A person is holding and aiming a long-barrelled firearm.
left side of the frame
The person is holding a vape device to their mouth with a visible cloud of vapor being exhaled.
right side of the frame, at the person’s mouth
Rules the customer wrote
Each card below is a policy an enterprise wrote in its own words, and the reason the platform returned when that policy fired.
The image shows a wild leopard walking on the porch of a residential home. frames 2–5 of 5
“Flag if there is any wild animal near home”
A bag or piece of luggage was identified in frame 1 that is not being held or stood next to by a person.
“Flag if there is any unattended luggage”
The worker was observed using a smartphone in a work setting in frame 1, which triggers the policy against smartphone use while working.
“Flag if worker use smartphone while working”
A vehicle was observed driving very close to or colliding with a person in frame 4, violating the workplace vehicle safety policy.
“Flag if any vehicle drives very near or hit to other people in workplace”
The policy text is reproduced exactly as the enterprises wrote it. Policies do not have to be well-formed English to compile, and nothing above was tidied up.
What video decides that a still cannot
A still image gives one moment of evidence. A video chunk gives a sequence, and
the reason the platform returns says which part of the sequence carried the
evidence: frames 2-5 of 5, in frame 1, in frame 4.
That is what makes the rules on this page writable at all. "Any unattended luggage" needs to know a bag stayed put while no one stood next to it. "A vehicle drives very near or hit to other people" is a trajectory, not a pose. Neither question can be settled from a single frame.
Where video safety sits
Video safety is the streaming surface. Industrial Safety builds on it for facility deployments, with prebuilt alerts and a camera fabric. Image Safety remains the right surface for a single still.