NEWVectors or files. Pick a path.Start →
    Search & Discovery
    7 min read
    Updated 2026-10-05

    How Do I Get an Alert When My Security Camera Sees Something Specific?

    To get an alert when a camera sees something specific, cut its live stream into short segments, skip the ones where nothing changed, describe and index the rest, and run a saved search against each new segment that notifies you when a match clears your threshold. Built-in camera detection covers people, vehicles and packages; this is for everything else, such as a red truck at the gate or someone without a hard hat.

    Security Cameras
    RTSP
    Alerts
    Video Monitoring
    Live Video

    How do I get an alert when my security camera sees something specific?



    Most cameras and recorders can already alert on a few fixed things: a person, a vehicle, a package, an animal. For anything more specific, such as a red truck at the loading dock, smoke, a person without a hard hat, or a delivery left at the wrong door, you need four steps: cut the camera's live stream into short segments, skip the segments where nothing changed, describe and index the ones that remain, and run a saved search against every new segment that notifies you when a match clears your threshold. Done this way, an alert can arrive about half a minute after the moment happens.

    What can built-in camera detection do, and where does it stop?



    ApproachWhat it can alert onWhere it stops
    Camera or app motion and object detectionMotion, people, vehicles, packages, animalsFixed categories; cannot tell a red truck from any truck
    Recorder or video-management analyticsZones, line crossing, loitering, licence plates on some systemsRules about where something moves, not what it looks like
    A saved search over indexed segmentsAnything you can describe in words or show with an example imageNeeds a pipeline and a threshold you tune per camera

    How does alerting on a description work?



    1. Capture. Pull the camera's RTSP stream and cut it into short segments, for example 10 seconds each. 2. Skip what did not change. Compare each segment with the last; a quiet parking lot at 3 a.m. is dropped before any model looks at it, which is where most of the cost of always-on cameras goes. 3. Index what remains. Each kept segment gets a visual embedding and, where it helps, a description and any text on screen, stored with its camera and capture time. 4. Run saved searches on arrival. "A red pickup truck at the gate" is a saved search. Each new segment is checked against it, and a match above your threshold sends an email, a webhook or a chat message with the clip. 5. Watch the watcher. A scheduled check that alerts when the feed stops arriving, so silence never reads as "nothing happened".

    How do I keep false alarms down?



    Tune the threshold per camera, because lighting and angle change what a good match scores. Require a match to hold across more than one segment before alerting. Describe exactly what you want, for example "a person in a yellow hard hat". Review a week of matches before turning alerts on.

    How do I do this with Mixpeek?



    The RTSP Live Monitoring template sets this up in one click: capture in 10-second segments, a gate that drops unchanged footage before extraction is paid for, searchable kept segments with their verdict scores as filterable fields, email alerts from retriever results, and a scheduled health watch. You add your camera afterwards as an ordinary storage connection, so the footage is read where it streams from.

    Measured on our own production runs: four live NYC DOT traffic cameras at once, about 35 seconds from capture to a searchable segment (2026-09-08), and 18 resources created by one click (2026-09-26).

    Related: how to find one specific moment in hours of footage and how Mixpeek alerts work.

    Frequently Asked Questions



    How do I get an alert when my security camera sees something specific?



    Cut the camera's stream into short segments, skip the ones where nothing changed, index the rest with a visual model, and run a saved search such as "a red truck at the gate" against every new segment. When a segment matches above your threshold, send an email, webhook or chat alert with the clip.

    Can my camera's own app do this?



    Camera apps alert on fixed categories such as people, vehicles, packages and animals. They cannot alert on a specific description like a particular vehicle, a missing hard hat or smoke in a certain area.

    How fast can the alert arrive?



    With short segments the alert can arrive within about a minute; in our runs a segment was searchable about 35 seconds after capture. Longer segments are cheaper and slower.

    Does it work with any IP camera?



    It works with cameras that publish an RTSP stream, which most IP cameras and recorders do.
    Managed Mixpeek

    Put multimodal search to work

    Connect a bucket and Mixpeek runs the whole multimodal search pipeline for you: extraction, indexing, and search over your own objects. No models to wire up, nothing to host.

    Start with Managed
    MVS · bring your own

    Already have vectors?

    Keep your embeddings on your own cloud and run dense, sparse, and BM25 search directly on object storage. From $25/mo.

    Start with MVS

    Run this on your own data

    Point Mixpeek at the storage you already have and search your video, images, audio, and documents the way this guide describes. Build starts at $25/mo for up to 1M vectors.

    Search your own archiveRead Docs