Violence detection in emergency departments
Violence detection in emergency departments is the visual detection of a physical altercation, on the waiting-room, triage and corridor cameras a hospital already has, raised as a critical alert with a clip to the security desk while it is happening. Emergency staff are assaulted more often than any other workers in the country, and the camera above the triage window records it for the review. Camzify raises it instead, to a person who verifies the clip and responds.
- Altercations raised live, not reviewed later
- On the waiting-room cameras already installed
- A person verifies the clip and responds

The security desk learns about it from the shout
An emergency department at two in the morning has a triage nurse, a registration clerk, a waiting room of people who have been waiting for hours, and a security officer who is somewhere in the building. The altercation goes from raised voices to a hand on a nurse in seconds. Security learns about it from a shout, a duress button, or a phone call from the desk, and the camera over the window records all of it.
The footage is reviewed afterwards. It establishes what happened and how long the nurse was alone with it, and it changes nothing about the minute.
Emergency physicians and nurses describe this as routine, and the published surveys below say the same. A department cannot put an officer at every window, and the cameras that already cover every window are watched by nobody.
“Dock door left unsecured after delivery — confirm and lock.”
Unacknowledged alerts escalate to the backup contact after the configured window.
Raised while it is happening, verified by a person
Aggression and fight detection watches the movement between individually tracked people on every enabled camera.
- Sustained pushing, striking or grappling between tracked people is the event; a crowded waiting room is not.
- The alert defaults to critical severity and carries the frame and a clip.
- An altercation involving three or more people is one tracked event, not three alerts.
- The person receiving it acknowledges it, and the record shows who and when.
Weapons detection is usually enabled on the same cameras, because a visibly held weapon in a waiting room is the same problem with a shorter clock. Both run at every hour; a notification window per camera limits when notifications go out, not when the detection runs, so the ambulance bay can route to the operations center overnight and to the desk by day. The notification settings cover channels, severity and escalation.
Afterwards, cloud backup holds the footage within the retention set for that camera, and incident investigation covers how the clip, the alert log and the acknowledgement are pulled together for a report. The broader violence and weapons detection page covers both detections in other settings.
Aggression and fight detection
Sustained aggressive movement between tracked people at the window, in the corridor or in the bay. A full waiting room does not match.
DetectionWeapons detection
A visibly held firearm or blade in the waiting room, raised as critical with a clip. Not concealed weapons.
DetectionMulti-object tracking
Each person followed as one track, so the fight model evaluates motion between people rather than movement in the scene.
DetectionNotifications and alerts
Critical by default, with channels and escalation per camera, so the desk, the charge nurse and the agency each get what they need.
- No altercation in progressCompliant
- Triage window staffedCompliant
- Security post occupiedNot compliant
- Camera view unobstructedCompliant
A failed item is resolved as Fixed or Pending before the round can close.
What a waiting-room round checks
The detection is continuous and does not wait for a round. The round adds the record: at each stop, no altercation was in progress, the window was staffed, the post was occupied, the camera could see. Repeated through the night, it shows the department was in the state its workplace-violence plan assumes, which is what a hospital is asked to show afterwards.
A failed item messages the person designated for that camera and stays Pending in the report until it is marked Fixed. On an automated round, the AI also raises a critical notification for a risk it sees at a stop that the checklist did not ask about. Guard notifications cover how a contracted officer receives and acknowledges it.
A check found Not Compliant captures a snapshot and messages the guard designated for that camera; on a manual round the operator chooses to send it, on an automated round it goes on its own. The item stays Pending until it is marked Fixed, which captures the after frame, and the report shows both.
Violence against emergency staff, in published figures
Figures from the Bureau of Labor Statistics, NIOSH, the American College of Emergency Physicians and the Emergency Nurses Association, quoted as the source states them. None of them is a Camzify figure.
- 91%
- In a January 2024 ACEP poll, more than nine in ten (91%) emergency physician respondents reported being threatened or attacked in the past year, and 71% believed violence in the emergency department was worse than the year before.
- Source: ACEP, April 2024
- 56 percent
- Of nearly 500 ENA members who answered an April 2024 survey, 56 percent said they had been verbally assaulted, threatened with violence or physically assaulted in the previous 30 days.
- Source: Emergency Nurses Association, April 2024
- 14.2 per 10,000
- In 2021-2022, health care and social assistance had 41,960 nonfatal workplace violence cases requiring days away from work, job restriction or transfer, 72.8 percent of all such cases in private industry, at an annualized rate of 14.2 per 10,000 full-time workers.
- Source: BLS, Workplace violence 2021-2022
- 76%
- Of workers who experienced trauma from nonfatal workplace violence in 2020, 76% worked in the healthcare and social assistance industry, and 22% required 31 or more days away from work to recover.
- Source: NIOSH, Violence and Work
What it will not do
The limits follow from what it is: a visual detection on the cameras that exist, verified by a person.
- It will not prevent an assault; it shortens the time between an altercation starting and a person in security knowing, and the response is theirs.
- It will not detect a threat that stays verbal; it has no audio, and most of what emergency staff report is verbal before it is physical.
- It will produce alerts on restraints, resuscitations and rough handling that a person, looking at the clip, will dismiss, and the workflow is built for that.
- It will not see a treatment bay or a room without a camera, and it does not ask for one.
- It will not identify anyone, and it does not link a clip to a patient or a staff record.
- It is not a workplace violence prevention program, and it makes no OSHA or Joint Commission compliance claim.
We do not publish detection rates, false-alert rates or response times for these models, and a page about assaults on nurses is the last place we would estimate one. The trust page sets out the policy.
Industries where this applies
Frequently asked questions
Ready to patrol your site 24/7?
Book a 15-minute demo and see a live patrol run on your own cameras.

