After an incident, footage is requested, the timeline is opened, and there’s nothing in that range. The disk isn’t empty, the server is up, the camera is working; the recording mode is event-based and motion detection didn’t trigger. When the camera records is a more fundamental decision than how many days of recordings are kept: the wrong mode results in the incident never being recorded, however large the storage. Below we compare continuous, event-based and scheduled recording in terms of evidential coverage, storage impact and the risk of missing events, and explain pre- and post-event durations, two-stream hybrid recording and choosing the mode by camera location. Calculating how many days of recording can be kept with the chosen mode is in the how many days does an NVR record article.
Recording modes
- Continuous recording: The stream is recorded without interruption. There’s no dependence on triggers.
- Motion-based recording: Recording starts when the camera or VMS detects a pixel change in the image. It’s the most common event-based method.
- Analytics event-based recording: Recording starts on a classified event such as a person, vehicle or line crossing. More selective than motion detection.
- Externally triggered recording: Recording starts on a signal from an external system, such as an alarm input, an access control event, a door contact or a cash drawer opening.
- Scheduled recording: Continuous at certain hours, event-based or off at other hours. For example, continuous outside working hours and event-based during working hours.
- Hybrid recording: A low-quality stream is recorded continuously, and a high-quality stream only at the time of an event.
Comparison table
| Mode | Evidential coverage | Storage | Risk of missing events |
|---|---|---|---|
| Continuous | Complete; context is recorded too | Highest | System failure only |
| Motion-based | Triggered periods | Depends on scene activity | Detection errors, threshold settings |
| Analytics event | Classified events | Low | Analytics misses (false negatives) |
| External trigger | Events tied to the trigger | Lowest | Signal or integration failure |
| Hybrid | Complete (low quality) + events (high quality) | Intermediate | Missed at high quality; context preserved |
The real distinction in the table is this: in continuous recording, the risk lies with the system itself (disk, server, network). In event-based recording, detection logic is added to that, and it’s a point of failure too. A silent one, at that. In event-based recording, there’s no way to tell afterwards the difference between “no event happened” and “the event wasn’t detected.”
What motion detection can miss
Motion detection compares the pixel difference between consecutive frames against a threshold. The logic is simple, and its weak points on site are well known:
- Slow or small movement: A person walking slowly in a far corner of the frame may not produce enough pixel change to exceed the threshold. Raising sensitivity, meanwhile, increases false triggers.
- Night noise and IR: Increased sensor noise in low light, insects glowing in IR light, rain and snow produce constant triggers. The operator lowers sensitivity to silence them, and real events start being missed too. This is the common chain: a flood of alarms at night, sensitivity turned down, a missed incident a few weeks later.
- Light changes: Cloud shadows, vehicle headlights or lighting switching on and off produce triggers; on some cameras, detection is temporarily suppressed during these changes.
- Detection zone: A zone narrowed to exclude a tree or a road with traffic can also exclude the area where the incident happens.
- Event transmission: An event detected on the camera reaches the VMS over the network; if event transmission is cut, recording doesn’t start. With server-side detection, the server has to fully decode the stream, which means processor load.
- A change in camera position: If the camera is pushed or a PTZ stays in a different position, the detection zones lose their meaning.
Analytics triggering that classifies people and vehicles reduces false alarms but brings its own miss rate. Factors affecting analytics accuracy are covered in the false alarms in video analytics article.
Pre- and post-event durations
Pre-event recording (pre-event buffer) adds the few seconds before the trigger to the recording; it shows the start of the event, that is, where the person entered the frame. Post-event recording keeps recording for a while after motion ends; it prevents the recording from fragmenting during short pauses.
| Field | Pre-event (starting point) | Post-event (starting point) |
|---|---|---|
| Office corridor, indoors | 5 s | 10–15 s |
| Entrances and doors | 10 s | 15–20 s |
| Perimeter and open areas | 10–15 s | 20–30 s |
| Externally triggered (cash drawer, door contact) | 10–20 s | According to the event duration |
These values are a starting point for tuning on site. One technical consequence of pre-event recording is often overlooked: for pre-event recording to be possible, the stream must already be arriving at the server before the trigger. The VMS receives the stream continuously and keeps the last few seconds in memory (on disk in some VMSs). That’s why event-based recording on the server saves storage but doesn’t reduce network traffic. If network traffic is to be reduced too, as in remote locations, the camera must upload at the time of the event or keep the event recording on the camera; this approach is explained in the edge recording and ANR article.
The hybrid recording approach
Hybrid recording, using the flexibility of dual-stream cameras, combines the context advantage of continuous recording with the storage advantage of event-based recording. The substream (for example 720p or lower) is recorded continuously; the main stream is recorded only at the time of an event, with pre- and post-event recording. In a review, the operator finds the event’s time range in the low-quality recording and watches the moment that needs detail from the high-quality recording. Context isn’t lost.
Worked example: On a 4 MP camera, let the main stream be 6 Mbit/s and the substream 0.5 Mbit/s, with motion lasting a fifth of the day (about 4.8 hours).
- Main stream continuous: 6 Mbit/s × 86,400 s = 518,400 Mbit ≈ 64.8 GB/day
- Substream continuous: 0.5 Mbit/s × 86,400 s = 43,200 Mbit ≈ 5.4 GB/day
- Main stream on events only: 6 Mbit/s × 17,280 s = 103,680 Mbit ≈ 13 GB/day
- Hybrid total: 5.4 + 13 ≈ 18.4 GB/day
In this example, hybrid recording gives about 3.5 times longer retention with the same storage than continuous recording of the main stream. The real ratio depends on the scene’s motion duration and the bitrate mode; with continuous recording using smart codecs and VBR, the difference narrows. The substream’s role and aspect ratio consistency in hybrid recording are explained in detail in the camera multi-stream management article. Hybrid recording must be supported by the NVR recording system or VMS; not every recorder can record two streams with different rules. Ask about this at the quotation stage.
Choosing the mode by camera type
| Camera location | Recommended mode | Reason |
|---|---|---|
| Checkout, cash and valuables areas | Continuous | Missing events unacceptable; context needed |
| Main entry and exit | Continuous or hybrid | Heavy, continuous motion; triggering is pointless |
| Production line, loading dock | Continuous (working hours) + events (outside working hours) | Process review needs context |
| Perimeter, open areas | Hybrid + analytics trigger | Long quiet periods, night noise |
| Archive, storeroom, technical room | Event-based + external trigger | Infrequent access, matching with door contacts |
| Office corridor | Scheduled or hybrid | Busy during working hours, sparse outside them |
| License plate recognition | Continuous or analytics event | The plate event is recorded separately |
The recording mode decision isn’t made separately from the retention period decision. With event-based recording, the same storage can last longer; but the retention period must also be justified in terms of personal data principles. Transmission of camera events to the VMS is standardized by the ONVIF event service; the ONVIF profiles show which profiles include this service. In installations mixing different manufacturers, test event integration separately for each camera.
Validation: checking for recording gaps
In enterprise camera systems, don’t assume event-based recording works; measure it regularly. The order of checks:
- Walk test: On every event-based camera, day and night, check from the edge of the detection zone and while walking slowly whether recording starts, and whether the pre-event recording includes the person entering the frame. A test done once at installation and then forgotten loses its validity when the season and lighting change.
- Timeline review: Look at a week’s distribution of events on the VMS timeline. A camera showing uninterrupted events all day is effectively on continuous recording and is probably being falsely triggered; a camera showing no events for days is either in a truly quiet area or its detection has broken.
- Event count alarm: The daily event count per camera falling outside its usual range (dropping to zero or a sudden spike) can be added to the monitoring system as an alarm. Zero events is an alarm too.
- External trigger test: If an alarm input or access control integration is used, physically operate the trigger and see recording start on the relevant cameras.
- Retention check: Compare the date of the oldest recording with the planned retention period. In event-based recording, storage filling faster than expected is an indirect sign of false triggering.
To plan recording modes and capacity together, see the camera recording storage and capacity planning service page.
Frequently asked questions
Does event-based recording also reduce network traffic?
If recording is done on the VMS or NVR, usually not. For the pre-event buffer and server-side motion detection, the camera stream keeps flowing to the server continuously; the savings are only in storage. To reduce network traffic, the camera must send the stream only at the time of an event or keep the event recording on the camera.
How long should the pre-event recording be?
As a starting value, 5–10 seconds covers most indoor scenarios. Perimeter and entrance cameras, where you need to see the person entering the frame and their behavior before the trigger, can use a longer pre-event recording. Whether the pre-event buffer is kept in memory or on disk, and its upper limit, vary by VMS.
Why does motion detection keep triggering at night?
In low light, image noise increases, and IR illumination makes insects and rain stand out; motion detection that looks at pixel changes counts these as motion. Adjusting sensitivity and threshold, narrowing the detection zone or using analytics that classify people/vehicles for triggering reduces the problem.
Which cameras must be on continuous recording?
Areas where missing an event is unacceptable and motion is continuous: checkouts and cash areas, main entry and exit points, critical production lines and areas whose recordings are later reviewed for context. Make the decision together with the area’s risk analysis and retention requirement.