Skip to content

Motion detection software that only fires on what moved

Your recorder already has motion detection, and that is the problem. It detects change, not objects, so rain, headlights and a moth on the lens all count as motion and a person at thirty metres sometimes does not.
Run it on a clip
  • Works with ONVIF and RTSP cameras
  • Person, vehicle, animal, package
  • No face recognition, ever

Interior, after hours

Fixed camera, low light, two subjects

00:00.0

Clip

Decoded in your browser. Your video is never uploaded.

6

Sets the luma delta threshold and the smallest region that counts.

0.4s

Kills the one frame flicker that no operator wants paged about.

Exclude zone

Put it over the road, the tree, the neighbour's window. Events inside it stop alerting.

The frame an event was found on shows up here.

00 00:00.0

Nothing selected yet

Hover a row in the event table to preview its frame, click it to hold the frame and jump the clip there.

Length
0.0s
Confidence
0%
Area
0.0%

0

Motion events found

0

Dropped by zone

0

Dropped, too short

0

Alerts you would get

# Start Length Class Conf. Area

Reading the clip. Events appear here as they are found.

01 Two different products wearing one name

Video motion detection and camera motion detection software are not the same thing.

What your recorder does

Frame differencing, included with the camera

The question it answers
How many pixels changed since the last frame?
What sets it off
Rain, snow, headlights, shadows, branches, insects, auto exposure shifts.
What it hands you
A notification, and a clip you have to open to find out what it was.
How you tune it
A sensitivity slider and a rectangle, retuned every season.
Where it fails
Night, weather, anything reflective, and any camera that can see a road.

What this does

Detection, then classification, then filtering

The question it answers
What class of object moved, for how long, and how sure are we?
What sets it off
A person, a vehicle, an animal or a package, above your minimum duration, outside your exclude zones.
What it hands you
An event row: start time, duration, class, confidence, motion area and one line saying what happened.
How you tune it
A polygon over what is not yours, a minimum duration, and which classes reach your phone.
Where it fails
Named openly on the how it works page, including distance, heavy occlusion and unusual camera angles.

Both columns are motion detection software. Only one of them can tell you that the thing at 02:14 was a person and the thing at 02:41 was rain, and that single sentence is the difference between an alert channel you read and one you muted in the first week.

02 Three filters, in this order

Most junk alerts are removed before the model is asked anything.

Classification is the expensive step, so it runs last, on what survived the two cheap steps in front of it.

01

Minimum duration

A raindrop crosses the frame in a fraction of a second. A person is present for several. Setting a floor of one to two seconds removes single frame flicker, most precipitation and almost all sensor noise, and it costs nothing to compute.

Removes: rain, snow, insects in flight, compression artefacts.

02

Exclude zones

Drop points on the first frame and close a polygon over the public road, the neighbour's window, the tree, the patch of floor a sunbeam crosses every afternoon. An event whose centroid falls inside is dropped before anything else happens.

Removes: passing traffic, footpaths, adjacent property, fixed reflections.

03

Classification

What survives gets a class: person, vehicle, animal, package, or other motion. You choose which classes reach a phone and which simply live in the timeline. Animal is usually the second one people mute.

Removes: everything that moved but was not the kind of thing you care about.

All three are live in the demo, and the counter under the video shows how many events each one removed. The detection itself is computed in your browser.

03 What an event looks like when it arrives

An event is six fields, and every one of them was measured.

Field Where the value comes from
Start time The timestamp of the first frame whose changed area passed the threshold.
Duration First frame to last frame of the grouped detection, in seconds.
Class The model's answer for the strongest keyframes of that event.
Confidence The model's own value, shown as it came back. Low numbers are shown low.
Motion area The share of the frame occupied by the changed region, as a percentage.
Description One line from the model about what it saw, in plain language.

The honest part

When the classification call fails or is rate limited, the event stays in the timeline marked unclassified, with its real motion data intact. There is no guessed label anywhere in this product, because a label you cannot trust is worse than no label at all.

Confidence is shown exactly as the model returned it. If it is 61 percent, the row says 61 percent, and you can decide what to do with a row like that.

More on camera analytics and the event timeline

04 Asked about motion detection software

Four questions, answered without hedging.

01

Do I have to replace my cameras?

No, and that is the whole point of this product. If a camera speaks ONVIF or RTSP it works, which covers effectively every business IP camera sold in the last decade, and most analogue cameras running into a hybrid recorder that can publish a stream. Details are on the RTSP software page.

02

Is this the same as the person detection in my new camera?

It is the same idea, done with a much larger model and with zone and duration filtering on top. On camera filters run on a small processor and are genuinely useful in good conditions. They struggle at distance, in rain and under infrared, which is where most false triggers come from.

03

Does it do face recognition?

No. It classifies object categories: person, vehicle, animal, package. It builds no face templates, does no gait analysis and does no re identification, and that is a product decision rather than a feature we have not got to yet. The security page states it in the terms your counsel will want.

04

What does it cost?

Between $8.32 and $12.25 per camera per month depending on tier, published in full on the pricing page. There is no free plan. The demo is the free part and it runs the entire detection loop with nothing held back.

Point it at the camera that wakes you up.

Run a clip through the demo first, in your browser, with nothing held back. When you want it on your own cameras, one email address starts it.