Adding and managing cameras

Camera configuration requires an administrator. Before adding anything, set the site defaults in Settings → Setup → Commissioning.

Add a camera

1. Open Add cameras from Cameras.
2. Under Camera passwords, add the camera logins used at the site. You don't need to know which login belongs to which camera.
3. Choose Scan network. EverNVR tries every saved password on each device it finds and shows the result beside it: unlocked (and with which login), or turned down every saved password. A password added after the scan is tried right away on the cameras still locked. Devices that answer with no camera service, such as routers, are listed separately.
4. If a camera doesn't appear, open Camera not found? to add it by IP address and optional ONVIF port, or by stream address for an RTSP-only source.
5. Untick any unlocked camera you don't want, then add the rest. EverNVR probes each camera's streams and lists what it found. A stream listed by ONVIF is not necessarily usable; distinguish advertised streams from verified ones.

Each new camera uses one connection. If you know a camera's simultaneous connection limit, enter it under Streams & connections so EverNVR can use its second stream. See camera connections. A known limit with no free connections beyond the reserved ones pauses the camera; do not reduce reservations merely to clear that warning.

The recorder's license covers a number of cameras. A camera past that number is refused before any of its connections is used, and the refusal says how many the license covers. Cameras already added keep recording.

The stream address a camera answers with over ONVIF is opened only if it is rtsp:// or rtsps://, carries no credentials of its own, and points at the camera's own address or a name that resolves to it. Anything else is refused with the address the camera offered. A camera that advertises an address it cannot be reached at, such as one behind address translation, is added with Add by stream address instead.

Camera settings

Select a camera in Cameras to open its settings:

Tab	Use it for
Overview	Recent camera snapshot, live state, selected streams, detector health, camera clock, and warnings
Setup	Name, Record and watch or Watch only, audio, live view mode, the camera's own events, deterrence outputs and talk, and removal
Streams & connections	Optional connection limit, reservations, recording stream, and shared viewing/detection stream
Detection	Confidence and size floors, detection schedule, and recognition floors
Recording	Recording policy, retention override, and tracking alignment
PTZ patrol	The presets a PTZ camera tours and how long it stays at each; see PTZ patrols. Cameras added over ONVIF only
Auto tracking	Whether a PTZ camera follows people or vehicles it detects, for how long, and where it goes afterwards; see PTZ auto tracking. Cameras added over ONVIF only

Configuration forms use an explicit Save button. Edit the fields, then select Save at the bottom of that form; changing a checkbox, radio button, or selection does not save it. Save before switching tabs or cameras.

Setup saves the name, recording mode, audio, live-view preference, and Camera's own events together. Streams & connections saves connection limits and stream choices together. Recording saves the policy, retention, and box alignment together.

Box alignment is how far detection trails the recording, so boxes over recorded clips and event stills line up with the footage. With Measured by the recorder, the default, the recorder remeasures when something moves and keeps every measurement; each clip uses the measurements taken nearest it, because the lag can change after a camera reconnects. The value shown beside Latest is the most recent measurement. Set by hand applies the value in Detection trails the recording by (ms) to all of the camera's footage. Detection settings and recognition floors have separate forms and Save buttons; choosing suggested floors fills the form for review before saving.

The Overview snapshot helps identify the selected camera and shows when its image was captured. It refreshes every 30 seconds while the page is visible and uses the existing camera feed, including privacy masks. If no recent image is available, it shows Preview unavailable.

Checking a different stream briefly pauses ingestion for that camera. Confirm recording resumes and repeat playback checks after a change.

Audio is off by default. When enabled, verified AAC is copied and verified G.711 is converted to AAC. Unsupported audio records video only with an explanation. The setting affects future recordings and does not change the camera encoder or open another camera connection.

The camera's own events

Many cameras run their own motion detection, line crossings, intrusion areas, people detection, and tamper alarms, and report them over ONVIF. Turn on Camera's own events on a camera's Setup tab to have EverNVR listen. It is off by default, and offered only for cameras added over ONVIF.

Listening uses no video connection, so it does not count against the camera's connection limit. EverNVR asks the camera to hold one event subscription for it, a minute at a time, and renews it while listening. When EverNVR stops, the camera drops the subscription within a minute. Nothing in the camera's configuration changes. Set up the camera's own detection, such as a tripwire or an intrusion area, in the camera's web interface.

Reports are sorted into Motion, Line crossing, Intrusion, Person, Tampering, and Other. Each report records whether something began, ended, or happened once. Read them under History → Camera events, or use them in a Camera reports rule (see zones and rules). When a camera first answers, it says how things stand. That is recorded as the starting state, not as something that happened.

These reports are the camera's judgment, not EverNVR's detection. They are not checked against the picture, and they do not appear as events, in the "something unusual" baseline, or in the morning briefing. The setting shows whether reports are arriving. After three failed attempts in a row, EverNVR raises a warning in Alarms and keeps retrying, less often each time. Recording and detection carry on regardless.

Camera clocks and PTZ

The recorder manages the clocks of the cameras attached to it. With Settings → This recorder → Camera clocks → Keep camera clocks correct on, which it is by default, it reads each ONVIF camera's clock every 15 minutes and again when a camera comes back online. When the time is more than a few seconds out, when the camera's picture shows a different time zone from the recorder's, when the camera keeps its own time instead of taking it from the recorder, or when it takes its time from another NTP server, the recorder sets all three: the time, its own time zone, and itself as the camera's NTP server. The picture time decides: a camera whose local time is within a minute of the recorder's is left alone whatever UTC time or offset it reports, since some cameras keep daylight saving by shifting their reported UTC time while the picture stays right. A camera already on NTP is asked which server it uses, and asked again at most once a day once it names the recorder; a camera that cannot say is left as it is. The recorder answers NTP on UDP port 123, so a camera without a battery-backed clock that restarts into the year 2000 puts itself right within a minute instead of waiting for the next check. A camera that refuses an NTP server keeps the corrected time and zone and is checked as before. A camera already showing the right offset keeps its own zone string, with daylight saving left on when its picture shows it. Each result is verified against the camera, by its clock or by what its picture time now reads, and every correction appears on History → Camera changes as Clock corrected, made by the recorder, with what was set and what the camera kept. One camera is checked at a time, a camera is corrected at most once every ten minutes, and a camera that did not keep the zone or NTP server from an earlier attempt is set again at most once a day; a routine correction of the time alone does not start that wait. A camera that refuses a request over its security timestamp is asked once more with the recorder's own clock, which is kept for that camera if it works, so PTZ, events and clock fixes still reach it. A correction of a minute or more, whether automatic or from Correct the clock, also reconnects the camera's streams, which costs a few seconds of recording: some cameras stop advancing the timestamps of a stream that was open while their clock jumped, which records footage that plays too fast until the stream reconnects. A camera without a battery-backed clock comes back reading 2000 after every power cut, so expect this once per outage. Turning the setting off stops the writes; the clock is still read on the same schedule and the drift still shown on the Overview tab.

Whatever the cause, and whether or not the setting is on, the recorder compares each stream's timestamps with its own clock. A stream whose timestamps run at less than about two thirds or more than one and a half times real time for a minute is reconnected, and if it happens again within 15 minutes the camera is recorded with the recorder's own arrival timestamps until the recorder restarts. Each time leaves a warning on History → Recorder health, already resolved because the recorder has acted on it, that names the camera, the stream and how far off its timestamps were, such as Lobby viewing stream timestamps ran at 0.25x real time; reconnected the stream. The check measures the browser copy of each H.264 or H.265 stream, so a camera whose live video is unavailable in the browser is not watched. A stream whose timestamps jump backwards, as when a camera's clock is set back under an open stream, is reconnected at once instead of leaving the live tile frozen, and a camera that has been delivering live video and then delivers none for 25 seconds is reconnected however busy its connection looks.

Two corrections are worth telling someone about, and both appear on History → Recorder health. A clock that was a day or more out leaves a warning already resolved (clock was 3 days out; corrected): it happened and it is fixed. A camera corrected three or more times in a day raises an open warning that it keeps drifting; its NTP source may be wrong, shown in recorder health under Settings → This recorder until a day passes without a correction. Warnings do not raise the problem banner.

When a camera's clock is more than a minute out, its Overview tab says so and offers an administrator Correct the clock, which sets the time now with the same safeguards. With the automatic correction on, that happens for a camera the recorder could not set. A camera on NTP that is still wrong after a correction has an NTP server it cannot reach; Switch to manual time and set the clock is then offered as a separate administrator action the recorder never takes on its own. Both are audited.

Operators and administrators can use PTZ on cameras they can access. Preset recall uses presets already stored on the camera; saving a preset requires an administrator. An RTSP-only camera has no ONVIF clock or PTZ controls.

On the Live page, a click on a PTZ camera's picture centers the camera on that point. When the camera has a calibrated or trained geometry (set up for privacy masks on Cameras → Privacy masks), EverNVR works out the pan and tilt from that geometry and sends the camera straight there, then checks where it came to rest and makes up to two small corrections if the point is still more than 3% of the frame off center. Cameras rarely describe their own field of view accurately, so a camera without a geometry, which has to be aimed in the camera's own terms, may need a second click to finish the job. A point the camera cannot physically reach, such as anything above the horizon of a dome that only tilts downward, is centered as far as the camera turns. Each click is audited with the error the model predicted and the error it achieved.

A PTZ camera placed on the site plan can also be sent to look at a point on the plan, by an operator from the plan page or by an action rule aiming it at what another camera saw.

The PTZ panel sits in a corner of the picture. Tap an arrow on the round stick to nudge the camera, or hold it to keep moving until you let go; the square in the middle stops the camera, and + and − zoom. Each preset stored on the camera is a chip under Presets: tap one to send the camera there. An administrator chooses Edit to manage presets: New saves the current position under a new name, and with a chip selected, Move here overwrites it with the current position and Delete removes it from the camera.

Drag the panel by its header to move it off whatever it covers, or tap the header to fold it down to a small PTZ pill. The browser remembers both for every camera. The gear opens the speed and nudge length used by the arrows; use Save there before trying the new values. Once the panel has been moved, the same form offers Put the panel back in its corner. On a phone held upright the panel sits below the video instead.

Deterrence outputs

Some cameras can switch something on: an alarm output over ONVIF (a strobe, a door strike, or a third-party siren wired to the camera's terminals), or the white light and siren built into an active-deterrence camera. On a camera's Setup tab, Deterrence lists what EverNVR found and lets an administrator Find outputs, or Look again once it has looked, and Test for 3 s beside each output. The recorder asks the camera rather than assuming from its model, and keeps the answer, so look again after changing the camera's wiring or account. Until someone has looked, the section reads Not looked for yet; a camera with nothing to switch reads Nothing to switch was found on this camera.

Each kind is reached the way its maker allows:

• ONVIF relay outputs through the camera's DeviceIO service, on any camera that lists them. One is named Alarm output; several are numbered.
• Dahua and Amcrest lights and sirens (White light and Siren) through the camera's own web interface, with the camera account EverNVR already has.
• Tapo sirens and spotlights (Spotlight and Siren) through the camera's local interface, which answers only to the Tapo app account, not the camera account. Enter the Tapo account password under Deterrence and select Save and look; it is kept encrypted with the camera's other credentials and checked against the camera before it is used, so a wrong password costs no sign-in attempts. Forget removes it.

Once found, an output appears in Live view under Deterrence beside the PTZ controls when an operator or administrator opens the camera: Light 30 s, Siren 10 s, or Output 10 s switches it on for that long, and while it is on the button reads Light off (or Siren off, Output off) to switch it off early. Each press is one deliberate act, never a toggle left behind. The same outputs are the Switch a light, siren or output action in action rules. An output switched on for a while, from one second to five minutes, is switched off on time even if the recorder restarts in between; if the camera does not answer, the recorder keeps trying for ten minutes past the time. Every switch, and every attempt the camera refused, appears on History → Camera changes as Output switched, with the rule's name when a rule asked and the camera's answer when it failed.

Talking through a camera

Many cameras have a speaker. EverNVR uses one that the camera offers through the ONVIF audio backchannel (ONVIF Profile T): the recorder asks each camera, once and in the background when it is added, whether its stream offers a channel for sound going to the camera, and keeps the answer. It does not go by what the camera claims elsewhere; some cameras claim an audio output they cannot be sent sound on. Deterrence on the Setup tab says whether talk is available and in which format, and Look again asks the camera again.

On the development bench, Amcrest cameras offer the backchannel and Tapo cameras do not; a Tapo camera's speaker answers only to TP-Link's own app. Operators and administrators talk from the live view; see talk through a camera.

Spoken messages

Settings → Spoken messages holds what a camera's speaker can say: a warning, a delivery instruction, a greeting. An administrator makes each one in either of two ways:

• Type a message: a name, up to 400 characters of text, and a Voice. The recorder speaks the text with Piper, once, when you save, and keeps the audio, so playing it later costs nothing. The first message in a voice downloads that voice, about 60 MB, from the Piper voices repository; each file is checked against a pinned checksum before it is used. The voices offered are ones their makers released into the public domain or under CC0.
• Record a message: a name, then Start recording and speak, up to a minute. Listen to the take, then Save message, Record again or Discard. Recording needs the browser's microphone, which needs a trusted HTTPS connection to the recorder.

Each saved message can be heard on the page and deleted. A message an action rule plays cannot be deleted until the rule no longer uses it; the refusal names the rule. Messages are kept in the recorder's database, so they are in its backups.

Operators play a message from the live view, beside Hold to talk, and an action rule plays one with Play a message (see action rules). Every play is on History → Camera changes as Talked through the speaker, with the message's name and, for a rule, the rule's.

Hardware evidence so far: an Amcrest active-deterrence camera's white light has been switched on and off from the recorder on the development bench, and its siren found but not yet sounded. The ONVIF relay path and the Tapo path are implemented with automated tests only; no camera with ONVIF relay outputs, and no Tapo camera with its app password entered, has been on the bench. Commission each output at the site and record the result.

PTZ patrols

A PTZ camera can patrol its presets. On Cameras → PTZ patrol, an administrator lists the presets to visit in order and how long the camera stays at each, from 10 seconds to an hour. The recorder runs the patrol itself, so it continues with no browser open, and it starts again from the first preset after a restart. A patrol with a single preset acts as a home position: the camera returns there once it has been left alone. Patrols can run always or only at certain times on certain days, like a detection schedule.

Any other PTZ command pauses the patrol. That includes an arrow press, a click on the picture, a preset recalled from the panel or by an action rule, and Stop. The patrol resumes with the next preset once the camera has been left alone for the time set on that page (2 minutes by default). On the Live page, a patrolling camera's tile reads Patrol or Patrol paused. The PTZ panel shows where the patrol is. Operators can select Pause patrol to keep the camera where it is until someone selects Resume patrol, and that pause survives a recorder restart. A patrol also waits while privacy masks are being calibrated or trained on that camera.

Each patrol step is a preset recall, so it follows the same rules. A camera with privacy masks that follow its position shows a blank picture while it moves. Detection relearns the new view instead of reporting the camera as moved. Zones are drawn on the picture, not on the scene, so check them at every stop on the patrol. Each step appears on History → Camera changes as a write made by the recorder. A preset the patrol visits, that an action rule recalls, or that auto tracking returns the camera to, cannot be deleted until the patrol, rule or auto tracking stops using it.

PTZ auto tracking

A PTZ camera can follow what its own detector sees. On Cameras → Auto tracking, an administrator sets up the follow and saves it:

Setting	What it does
Auto tracking is on	Switches the follow on or off without losing the rest of the settings. A camera with nothing saved says Auto tracking is not set up on this camera above the form and Not saved beside Save; nothing follows until it is saved
People / Vehicles	What to follow; choose one or both. Vehicles covers cars, trucks, buses, motorcycles and bicycles
Follow one object for at most (seconds)	The time limit on a single follow, 10 seconds to 10 minutes; 2 minutes unless changed
Return to preset	Where the camera goes after a follow ends on its own: Stay where it stopped, or one of the camera's presets, after (seconds) a wait of up to 10 minutes (10 seconds unless changed)
After someone else moves the camera, wait (seconds)	How long auto tracking stays off after an operator drives the camera, 10 seconds to an hour; 1 minute unless changed

Delete removes the settings altogether; the camera stays where it is. Auto tracking needs two things from the camera, and the tab says which one is missing. Its PTZ geometry must be calibrated or trained on the Privacy masks tab, because the recorder works out where to point the camera from it. The camera must also accept absolute positions over ONVIF. Until both are there the settings can be saved switched off but not switched on.

When the detector sees a person or vehicle of a chosen kind travel across the picture, the recorder turns the camera to keep it in the middle. The camera moves in quick steps rather than continuously: each step is one move, aimed slightly ahead of the object so it walks through the middle of the view, and the next step is sent only once the picture has settled. Something that was standing in view when detection learned the picture, such as a parked car, is not followed until it moves, and neither is a detection that comes and goes in one place. The recorder lets the object go once it has been out of sight for five seconds, or when the time limit is reached. Auto tracking also waits while privacy masks are being calibrated or trained on that camera.

A person being followed rarely walks at one steady pace, so the follow allows for stops, starts and turns:

• After a step lands, the camera waits to see the object move again before it leads it. Someone who stopped while the picture was turning is centered where they stand rather than chased ahead; the next step leads them only once the new view shows them walking.
• If the object is not in the new view about a second after a step lands, the camera goes straight back to where the object was last seen, once, rather than waiting out the five seconds. Someone who stopped short or turned back during the step is usually there.
• Someone who stops after rushing past, or who sets off from standing, is looked for both where their pace says they should be and where they were last seen, so a follow is not lost with the person still in the picture.
• A box of the wrong size is not the object. At the same zoom the followed person or vehicle stays about the same size from one picture to the next, allowing for walking nearer or further; a much smaller or larger detection near where the object should be, such as a false detection on a wall, is ignored instead of taking over the follow.

While the camera follows, detection still finds the object in every picture but records no visits, zone events or tamper reports. The visit that started the follow ends when the follow begins. Once the follow ends and the camera settles, detection learns the new view as it does after any other move. On the Live page the camera's tile reads Following and the kind of object, such as Following person or Following car, and a box is drawn around the object. The PTZ panel says what is being followed and since when, and offers Stop following. The Auto tracking tab shows the same state and, below it, how the last follow went: what was followed, for how long, how many moves it took and why it ended.

Anyone else's PTZ command ends a follow at once. An operator's arrow, click, preset, Stop or Stop following also pauses auto tracking for the wait set on the tab, and the panel then reads Auto tracking paused with the time it resumes if the camera is left alone; when that time passes the panel goes back to Auto tracking · watching by itself. A follow that reaches its time limit pauses auto tracking for the same wait, so a person standing in view is not followed again straight away. The pause is not kept across a recorder restart. An action rule that recalls a preset or aims the camera at another camera's detection ends a follow without pausing auto tracking, so the camera can take up whatever it finds where the rule pointed it. A follow pauses the camera's patrol and cancels a return that a rule booked. After a follow that lost its object or ran out of time, and that actually moved the camera, the camera goes back to the Return to preset after the wait, unless somebody drives it first. Each follow appears once on History → Camera changes as a write made by the recorder, with what was followed, for how long, how many moves it took and why it ended; the return afterwards is a second write. The recorder's service log also carries one line per step, with the step's size in degrees, where the object was and how fast it was moving, where it was expected to be in the picture when the step landed, and whether the step was a return to where the object was last seen, so a follow that went wrong on site can be read back afterwards with journalctl -u evernvr.

How closely a camera keeps up depends on how long it takes to answer a move. On the camera this was developed with, the picture starts to turn about one to two seconds after the command, depending on the stream, so a fast object close to the camera can leave the picture between steps and is picked up again when the next step lands. A camera with privacy masks that follow its position shows a blank picture during each step if the masks are set to blank while moving; check the masks during a follow at commissioning.

Hardware evidence so far: on one camera model on the development bench, a parked vehicle pushed off center was brought back to the middle in one move and held there, and a person has been followed walking through the bench scene. Those walks showed the follow going wrong when the person stopped, turned back, walked close to the lens, or was replaced by a small false detection on a wall; the stop, start, turn and size handling described above came from them and was tuned in simulation against that camera's measured response and recordings of those walks. No installed recorder has run auto tracking. Commission it at each site, watch a real walk-through that includes a stop and a turn, and record the result.

After adoption, check both EverNVR's playback and the incumbent recorder's footage. Record the outcome in Settings → Setup → Commissioning. See the installation guide for the full procedure.