Skip to content
2026.08.14 Cost, 8 minute read

Free motion detection software, and where it breaks

What the free tier of every tool gives you, and the exact camera count and lighting condition where it stops being enough.

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.

Free motion detection software is a real answer for a lot of people, and anybody selling a paid product who tells you otherwise is worth ignoring. Plenty of sites run for years on the detection that came with the recorder, and never need anything else.

This page is about the boundary. What the free options genuinely do, and the four specific conditions under which each one stops being enough. If none of those four conditions apply to you, stay free, and the setup advice below will still make your alerts better than they are today.

What free actually means in this category

Three different things wear the word free, and they behave nothing alike.

  • Included. The detection in your NVR or camera. Costs nothing extra, works out of the box, detects pixel change.
  • Free software, your hardware. Frigate, ZoneMinder, MotionEye. No licence fee, but the machine, the accelerator and the maintenance are yours.
  • Free tier of a paid product. One camera, or a few days of history, designed to end. Useful for evaluating, not for running a site.

The first two are honest and can last forever. The third is a sales mechanism, which is fine as long as you read it as one.

The free options worth your time

The detection in your recorder or camera

Every Hikvision, Dahua, Reolink, Amcrest and UniFi setup has this. Draw a region, set a sensitivity, get notified. Newer cameras include an on device person or vehicle filter, which is a genuine improvement over raw pixel change and is often switched off by default. Turn it on before you conclude anything about your cameras.

Frigate

The best free option by a distance if you can host it. Real object detection on live streams, proper zones, and a configuration you can keep in version control. Wants a machine and an accelerator. The full honest total is in open source video analytics and the cost nobody prices in.

MotionEye and motion

Lightweight, will run on a small board, excellent for one or two cameras and for a webcam. It is frame differencing rather than object detection, so it inherits every weakness of class one detection while giving you far more control over the thresholds. For a single webcam pointed at a door it is entirely sufficient, and there is a setup walkthrough in motion detection software for a webcam.

Free tiers of cloud camera apps

Usually one camera, a short retention window, and detection that is better than your recorder's. Read what happens at camera two before you build anything on it.

The four breaking points

In our experience of this problem, and in the pattern of what people are trying to fix when they arrive here, free stops being enough at one of exactly four moments.

1. Camera count, somewhere around six to eight

One camera generating four junk alerts a night is a mild annoyance. Eight cameras generating four each is thirty two, which is the number at which a human being turns the channel off. It is not a linear decline, it is a cliff, and the cliff is roughly where per camera tuning stops being something anybody is willing to do on a Sunday.

2. Infrared at night

This is the big one and it is physics, not software. Under an infrared lamp, rain, snow, insects and spider webs are lit far more brightly than anything at distance, and a frame differencing detector correctly reports enormous change. No threshold setting fixes it, because turning the sensitivity down far enough to ignore a moth also ignores a person in dark clothing at thirty metres. Only classification separates them. The full physics is in why your security camera keeps sending false triggers.

3. A public road, a footpath or a neighbour in frame

Detection regions are rectangles in most recorder interfaces, and roads are rarely rectangular. If the only way to exclude the road is to exclude half your own forecourt, you have lost the coverage you bought the camera for. Polygon exclude zones plus class filtering solve this properly, and you can watch it happen on the demo: draw a zone over the road, and the events whose centroid falls inside it drop out of the timeline while the counter ticks down.

4. Somebody other than you needs the answer

The moment a second person needs access, or an insurer wants a retention policy, or IT asks who can view which clip, free stops being about detection quality at all. A box in a cupboard with one administrator password has no roles, no audit log and no data processing agreement, and nobody in procurement is going to accept a screenshot.

Getting the most out of free, in one afternoon

  1. Turn on the on camera person or vehicle filter if your model has one. This is the single biggest free improvement available and it is frequently off.
  2. Shrink the detection region to the ground you care about. Sky, road and neighbouring property out.
  3. Raise the minimum object size until the moth stops qualifying, then check a person at your longest distance still does.
  4. Set a minimum duration of one to two seconds if the interface offers it. This alone removes most flicker and rain.
  5. Move the infrared lamp away from the lens if it is mounted on the camera body, or use external illumination. Fewer close bright insects, fewer alerts.
  6. Clean the dome. Spider webs across a lens are one of the most common causes of a camera that alerts every night at the same time.

If that afternoon gets your nights quiet, you have solved the problem for nothing, which is the right outcome and we would rather you had it than not.

The point where paying is cheaper than free

Do this arithmetic honestly. Count the minutes a week you spend dismissing alerts, add the minutes spent scrubbing footage after an incident, add the hours maintaining anything self hosted, and put your own rate on it. Then compare against the published price: $49 a month for four cameras, $149 for fifteen, which is $9.93 a camera. Paid stops being an indulgence at the point where the time costs more than the subscription, and for most sites past about eight cameras it already does.

Why there is no free tier of this product

Because a free tier of a detection product is either useless or dishonest. One camera and a few days of history does not tell you whether the alert storm on your site stops, which is the only question you actually have. So instead of a limited plan, the whole detection loop is in the demo: bundled clips or your own footage, real frame differencing computed in your browser, real classification on the events it finds, nothing held back and no account required. Every listed plan is paid, and the demo is the free part.

Written by the people building Motiondetection. Prices quoted for other products are those vendors' own published positioning at the time of writing, not quotes we were given, and they change. Check them yourself before you decide anything.

The free part here is the whole detection loop.

Run it on a bundled clip or your own footage, with no account and nothing held back. The paid part is doing it continuously on your own cameras.