Home security monitoring systems are a duty-cycle problem. The SERP sells them as a price-tier problem.
Every top result for home security monitoring systems compares ADT, Vivint, SimpliSafe, Ring Professional, and Nest Aware on the same axes: monthly price, whether a human is in the loop, and contract length. None of them publish the number that actually decides whether your monitor ever sees the event: frames inferenced per day. This page prints that number for every common pipeline, shows the log artifact that proves which one you bought, and explains why two systems that both advertise 24/7 monitoring can differ by three orders of magnitude in time spent looking.
Audit the duty cycle on your own camerasTwo pipelines share one phrase, and the phrase is doing a lot of work
The phrase home security monitoring systems has been folded around two completely different architectures. The older shape is event triggered: a camera or a door sensor holds the pipeline dormant until a trigger fires, records a short clip, and either uploads that clip to a cloud inference endpoint or to a human at a central station. Everything Ring Professional, Nest Aware, ADT, Vivint, and most SimpliSafe camera tiers do follows this shape. The younger shape is continuous frame: an inference loop is running against the live video whether or not anything is happening. On residential deployments the continuous shape usually shows up as an NVR with embedded analytics, or as an HDMI-tap edge unit sitting between an existing DVR and the monitor that already shows the multiview.
These two architectures are sold under the same label. That is the source of the confusion. Both describe themselves as 24/7 home security monitoring. Both list person detection, vehicle detection, and proactive alerts on the feature page. Both quote a monthly price. None of them print the one number that separates them: how many frames the pipeline actually runs a forward pass on, in a day, from a camera you have installed. That number decides whether your system notices a stationary loiterer, a slow approach, a vehicle idling outside a fence, or a package sitting in a vestibule.
This guide prints the number.
The duty-cycle numbers, stated plainly
Four anchor numbers describe the inference budget of a continuous-frame home security monitoring pipeline on an HDMI tap. The comparison number that matters is not the monthly price; it is the ratio between these numbers and the frames an event-gated system runs.
The lower bound assumes 24 fps, a conservative reading of the hdmi_capture log line the edge unit prints each tick. At the upper bound of 30 fps the same 24 hour window runs 0 forward passes per composite, which maps to up to 25 cameras inside that composite. An event-gated camera, by contrast, is usually between zero and twenty thousand frames for the same day.
Where the two home security monitoring pipelines diverge
The duty-cycle ladder, every common home monitoring pipeline
The table below is the ladder a buyer should read before comparing monthly prices. Each row describes what the pipeline does between events, how many frames it actually inferences in a day, and whether stationary-object dwell detection is architecturally possible for that row. Dwell is the canary detection class; if the pipeline cannot see a person who is not moving, it cannot do most of the advanced analytics the marketing page implies.
| Monitoring pipeline | Trigger | Frames / day / cam | Dwell? | Offline behaviour |
|---|---|---|---|---|
| Consumer Wi-Fi cam, self-monitored | PIR / pixel-delta, 10-30 s clip | ~2k - 20k | No | Clip lost if uplink times out |
| Ring Professional / Nest Aware | PIR / pixel-delta, 10-30 s clip uploaded to cloud | ~2k - 20k per cam | No | Clip lost if uplink down long enough |
| ADT / Vivint central station | Sensor fire + clip handoff to human | ~5k - 30k per cam (event-gated) | No | Cellular fallback helps, still event-gated |
| Cloud VMS with motion analytics | On-stream motion mask, short sliding clip | ~50k - 250k per cam | Partial (sliding window) | Uplink required for inference |
| On-site NVR + embedded analytics | Per-stream RTSP decode, continuous | ~1.3M - 2.6M per cam (but 1 pass per stream) | Yes | Local, but cost scales per camera |
| HDMI-tap edge unit (Cyrano) | HDMI composite, continuous | ~1.3M - 2.6M per composite, up to 25 tiles | Yes | Queue locally, deliver when link returns |
Rows one through four are event gated. A stationary loiterer that outlives the PIR cooldown or the sliding-clip window is invisible to them. Rows five and six are continuous. The difference is not about feature lists; it is about whether the pipeline has any frames to run the feature on in the first place.
The compute budget, in twenty-four lines of pseudocode
Two ways to run the phrase home security monitoring. The left hand shape is what event-triggered products do: sleep until a trigger fires, then burst a short clip into a model. The right hand shape is what a continuous-frame edge unit does: read the composite the DVR is already drawing, run one forward pass per tick, filter downstream. The number of cameras inside each composite is decided by the DVR multiview, which means a single forward pass can cover up to twenty-five tiles.
Event-triggered vs continuous-frame home security monitoring
# Event-triggered home security monitoring
# Camera holds the pipeline dormant until a local trigger fires.
while True:
trigger = pir_or_pixel_delta() # dormant until fire
if not trigger:
sleep(1 / 30) # not inference, just wake poll
continue
clip = record_clip(duration=30) # bounded window, 10-30 s
upload_to_cloud(clip) # latency governed by uplink
boxes = cloud_model(clip) # one inference burst per event
if human_review_tier:
notify_operator(clip, boxes) # ADT / Vivint central station
notify_owner(boxes) # push to phone
# Frames inferenced per day: ~2k to ~20k (event-gated)
# Coverage between events: zero
# Stationary-object dwell: invisible by constructionWhat a continuous tick actually prints
The log slice below is one composite frame on a sixteen camera residential or small-multifamily install, with the 4x4-std layout active. A continuous-frame pipeline produces a readable line per tick whether or not anything is happening. If a vendor will not show you equivalent output on their own hardware, that is the signal that the pipeline is event gated.
The note line is the one a pixel-delta or PIR system cannot produce. motion_delta=0.3% means the person has been almost perfectly still for 41 seconds. An event-gated pipeline would have already timed out and returned to dormant.
Detection classes mapped to duty cycle
The cards below sort the advertised detection classes of a home security monitoring system by whether they can exist in an event gated pipeline at all. Classes that need only the first few seconds of motion survive event triggers. Classes that need to observe absence of motion, or long continuous tracking, do not.
Fast motion alert at a door
Fires inside the first second of any trigger, so event-gated pipelines handle this fine. The alert is usually the only thing consumer Wi-Fi cameras actually deliver well.
Package drop and dwell
Needs to observe a small object arriving and then remaining still. The dwell half dies in an event-gated window. Continuous-frame handles it because the object never stops being in the inference input.
Stationary loitering
Needs continuous tracking of a person who has stopped moving. Architecturally invisible to PIR and pixel-delta pipelines. First-class on continuous-frame.
Tailgate at a vestibule
Needs two silhouettes in the same frame. Survives event-gated only if the second silhouette is still in frame inside the clip window. Continuous-frame handles the case where the tailgater lingers.
Vehicle idling outside a perimeter
First arrival is a trigger event. Continued idling for minutes is silence to a PIR pipeline. Continuous-frame tracks it across the whole dwell.
Fence-line approach at crawl speed
Someone moving a centimetre per second never trips pixel-delta. A continuous pipeline sees the per-frame displacement and fires on zone entry regardless of speed.
“The question you ask a home security monitoring vendor is not 'is there a human in the loop' or 'what is the monthly price'. It is 'how many frames of my camera do you actually run a forward pass on in a day, and can you print me the log'. Everything else is marketing language around that number.”
Cyrano field notes, duty-cycle audit
How to audit the duty cycle of any home security monitoring system
Four checks, in order. Each check is answerable by a short written reply from a vendor or by a sixty second reading on your own dashboard. If the vendor cannot answer even the first two in plain terms, the pipeline is probably event gated and the marketing page is probably overstating what it sees.
Ask for a per-tick log from production hardware
A continuous pipeline prints a line per tick at the fps ceiling of the input. An event-gated pipeline is silent between triggers. Ask for a 60 second window. On a Cyrano edge unit the first line of the trace is hdmi_capture frame_id=0x.... res=1920x1080 fps=24, and every subsequent tick prints the same shape.
Ask for the fps field on each tick
If the number is present and steady, the pipeline is continuous. If the trace only shows a handful of lines per trigger window, the pipeline is event gated regardless of what the monthly price page says.
Ask what gates inference
Three answers are possible: PIR, pixel-delta, always-on. Only the third answer is continuous. PIR and pixel-delta both mean the monitor is sleeping most of the time and will miss stationary events by construction.
Ask for the 24 hour inference count
A continuous-frame edge unit on one composite reports 1.3M to 2.6M forward passes in a day. An event-gated camera typically reports 2k to 20k. The ratio is the duty-cycle gap. If the vendor refuses to print this number, you are buying the marketing language, not the duty cycle.
Continuous-frame vs event-triggered, feature by feature
The comparison table below maps the detection and behaviour features most home security monitoring systems advertise against the two pipeline shapes. The left column is what survives on a continuous-frame HDMI tap. The right column is what survives on an event-triggered cloud pipeline. Both are legitimate products; they are not interchangeable.
| Feature | Event-triggered (PIR or pixel-delta clip) | Continuous-frame (HDMI tap) |
|---|---|---|
| Fast motion alert at a doorbell | Yes, once the PIR or pixel-delta trips | Yes, first frame after the person appears |
| Stationary loiterer beyond PIR cooldown | No, pipeline has returned to dormant | Yes, still being inferenced every tick |
| Package drop plus dwell detection | Partial, loses the dwell half after the clip | Yes, same pipeline sees arrival and stillness |
| Vehicle idling outside a fence for ten minutes | No, only the first clip window is seen | Yes, dwell timer is first-class |
| Fence-line approach at crawl speed | No, sub-threshold motion never trips the trigger | Yes, per-frame displacement is tracked |
| Alerts in the 0 to 5 second range from event to phone | Possible on good networks, depends on uplink | Typical, because inference already ran on the frame |
| Works with internet offline | No, the cloud is the pipeline | Yes, inference continues, delivery queues locally |
| Covers 25 cameras on one monthly fee | Usually per-camera pricing | Yes, one composite covers up to 25 tiles |
| Raw frames stay on property | No, clips are uploaded to cloud inference | Yes, only filtered payloads leave |
Pre-buy checklist for any home security monitoring vendor
- Ask for a 60 second per-tick log trace from production hardware.
- Ask for the fps field on the tick.
- Ask whether inference is gated by PIR, pixel-delta, or always-on.
- Ask for the 24 hour forward-pass count on a single camera.
- Ask whether frames leave the property or stay on-site for inference.
- Ask whether a stationary loiterer beyond 60 seconds would fire an alert.
- Ask what happens to inference during an internet outage.
- If any answer is a price tier or a feature bullet, ask again in plainer language.
DVRs and NVRs a continuous-frame home monitor can read from
The layout classifier on the edge unit matches on tile geometry and corner anchors rather than on DVR firmware, so the list above is illustrative rather than exhaustive. Any recorder that drives a standard 1920x1080 HDMI multiview is a valid input for a continuous-frame home security monitoring system.
See the duty cycle of your own home monitoring on a live call
15 minutes. We attach a Cyrano edge unit to a production DVR, print a 60 second per-tick log, and you read the fps field, the layout_id, the mask file, and the 24 hour forward-pass count in real time.
Book a call →Home Security Monitoring Systems: Frequently Asked Questions
What is a home security monitoring system actually doing between events?
It depends entirely on which of two pipelines you bought. An event-triggered cloud system (Ring Professional, Nest Aware, most SimpliSafe and Vivint camera tiers) is doing nothing between events. The camera's on-board motion sensor or pixel-difference check holds the system dormant; only after a trigger fires does the service upload a short clip and start monitoring it. A continuous-frame system (an on-site edge unit watching a DVR or NVR output) is running a forward pass on the live composite every tick, 15 to 30 times per second, whether or not anything is happening. The difference in time spent looking is the gap the SERP hides.
How many frames of video does a typical home security monitoring system actually inference per day?
Continuous-frame monitoring on an HDMI-tap edge unit runs one forward pass per composite frame at roughly 15 to 30 fps. That is 1.3 million to 2.6 million frames per unit per day. Event-triggered cloud monitoring only inferences the 10 to 30 second clips uploaded after motion triggers. On a single busy front door that is somewhere between 2,000 and 20,000 frames per day. On a quiet back-yard camera it can be zero. Same 'monitoring' label on the box, two to three orders of magnitude gap in frames actually seen.
Why does this duty-cycle difference matter for detection quality?
Every event that is not preceded by a motion trigger is invisible to an event-triggered system. A stationary loiterer who holds still for a minute past the PIR cooldown drops off the pipeline entirely. A vehicle idling outside a fence is seen for the first 10 seconds of its arrival, then lost. A continuous-frame pipeline never stops looking, so dwell timers, zone re-entry, and stationary-object detection all work. Those are the detection classes that SERP comparisons call 'advanced analytics' without saying they require continuous input to exist.
Is 'professional monitoring' the same as continuous monitoring?
No. Professional monitoring refers to whether a human at a call center reviews alerts, not to whether the monitor is always looking. ADT, Vivint, and similar services are almost always event-gated: the camera or sensor trips, an alert goes to the central station, a human opens the clip. The human never sees what happened between triggers because the system never recorded continuous inference on it. Professional monitoring and continuous monitoring are orthogonal axes, and the SERP conflates them.
What does a log line from a continuous-frame home security monitoring system actually look like?
On a Cyrano edge unit watching a DVR in 4x4-std layout, every tick prints a line like: hdmi_capture frame_id=0xa1f02 res=1920x1080 fps=24, followed by layout_classify result=4x4-std, layout_cache hit=4x4-std mask_file=/var/lib/cyrano/layouts/4x4-std.mask, forward_pass model=person-v7 tiles_seen=16 pass=1 latency_ms=22, and so on. Every frame the model runs emits a log line. If you ask a vendor for equivalent output and they cannot show you a per-tick trace from production hardware, the pipeline is not continuous.
How should I price-compare a continuous-frame monitor against an event-triggered one?
Do not compare monthly cost alone. Compare cost per frame inferred. A Ring Professional plan at roughly 20 USD per month on a single camera inferences on order of tens of thousands of event-triggered frames per month. A Cyrano unit at 200 USD per month covers up to 25 cameras in one composite and inferences on order of 40 to 70 million frames per month (1.5M frames/day x 30 days x up to 25 cams implied by the multiview). The per-frame cost of continuous inference is orders of magnitude lower, even though the monthly label looks larger. That is the arithmetic the SERP leaves out.
Do I need to replace my existing cameras to get a continuous-frame home security monitoring system?
No, if your recorder outputs a standard HDMI multiview to a guard or home monitor. The edge unit sits on the HDMI line between the DVR/NVR and the monitor and reads whichever composite the recorder is already drawing. Any recorder from Hikvision, Dahua, Lorex, Swann, Uniview, Annke, Reolink, Night Owl, Amcrest, or the long tail of rebranded DVR hardware that drives a 1920x1080 multiview is a valid input. The cameras themselves never move; no rewiring, no camera replacement.
What happens when my internet is out? Does monitoring stop?
On a continuous-frame edge unit: inference keeps running on the live composite, the event queue fills locally, and delivery pauses at the network edge until the link comes back. Nothing during the outage is lost to detection; only the alert delivery is delayed. On most event-triggered cloud systems: the outage is the system, because the camera uploads clips to the cloud for inference. If the link is down long enough to drop the trigger into the on-device buffer and then evict it, the event is permanently unseen. This is the reason privacy-sensitive and bandwidth-poor homes gravitate to on-site inference.
Where does the inference run on the specific case of Cyrano?
On the Cyrano edge unit itself, which sits on the HDMI line between the DVR or NVR and the guard or homeowner monitor. Model weights, layout cache, per-layout mask files, the filter stack, and the event queue all live on local storage. The only thing that leaves the unit is an already-filtered event payload (thumbnail, metadata, camera name, zone, dwell seconds, layout_id, latency_ms) sent over the customer's network to their WhatsApp or SMS endpoint. Raw frames never leave the property, which is also why the per-frame inference cost is bounded by local hardware rather than a cloud GPU bill.
How do I audit whether my current home security monitoring system is actually continuous?
Ask the vendor for four artifacts: first, a live log trace showing one line per tick for a 60 second window; second, the fps field on each tick; third, the trigger condition that gates inference (PIR, pixel-delta, always-on); fourth, the retention of raw inference records for a 24 hour window. If the pipeline is event-triggered, the trace will show activity only during motion windows. If it is continuous, the trace prints uninterrupted at 15 to 30 fps. There is no vendor-specific language in this test; it is the same four questions regardless of brand.
Comments (••)
Leave a comment to see what others are saying.Public and anonymous. No signup.