Smart Filtering
AI Motion Classification
Person, vehicle, package, animal, or other. Halstead subscribes you only to the categories you care about. Person at 3 a.m.? Notify. Squirrel at 3 a.m.? Skip.
What it is
AI motion classification is a real-time computer vision system that identifies what kind of object triggered the motion — distinguishing humans from cars from packages from animals from environmental motion (wind, shadows, lights). Halstead applies this AI to every motion event before deciding whether to notify you, so you get fewer but more relevant alerts.
How it works
Step by step.
Motion detected
A camera detects motion in its field of view. The video frame is processed locally on the Halstead Hub for low latency.
AI classifies the object
A computer vision model identifies what's in the frame: person, vehicle, package, animal, or other. The classification typically completes in under 200 milliseconds.
System checks your subscription preferences
Each motion category has its own notification preference per camera, per time of day. You can subscribe to person alerts only, or person + package, or all categories — whatever fits your needs.
Notification fires (or doesn't)
If the classified object matches your subscription, you get a notification with a snippet of video and the classification label. If it doesn't match, the event is logged in your timeline (searchable later) but no notification interrupts your day.
Why it matters
Without AI classification, motion detection becomes useless because you get notifications for every car passing, every shadow, every animal. Most users disable motion alerts within weeks. With AI classification, your alerts become trustworthy — you only get notified for events you actually care about, which means you actually pay attention when alerts come through.
Technical details
Halstead uses an on-device computer vision model running on the Halstead Hub for the fastest possible classification (typically 100-200ms). Cloud-based fallback handles edge cases. The model is trained on millions of labeled images from residential security contexts and continues to improve through ongoing model updates. Classifications include: person (with rough size estimation), vehicle (car/truck/motorcycle), package (placed and removed events), animals (dogs, cats, deer, raccoons, squirrels, etc.), and other (unclassified motion).
How Halstead compares
Most consumer security platforms offer some level of object detection (Ring's 'Smart Alerts' feature, Nest's 'Familiar Faces') but typically require their highest subscription tier and don't differentiate as many categories. Halstead's AI is included on the Premium tier and above as a default rather than a paid add-on.
Common questions
How accurate is the AI classification?+
Person and vehicle classification typically achieves 95%+ accuracy in normal lighting. Package detection (placement vs removal) is more nuanced — typically 85-92% accurate depending on lighting and camera angle. Animal classification varies by species. Halstead's model is updated regularly, so accuracy improves over time.
Does this work in the dark?+
Yes. Halstead's cameras have IR night vision that produces a clear black-and-white image at night, and the AI model is trained on both color and IR night vision images. Detection accuracy in low light is slightly lower than daytime but still strong for the categories that matter most.
Can the AI recognize my specific family members?+
Yes — through the Family Check-Ins feature. Pre-register household members at install with consent, and the AI can distinguish family members from strangers. This enables features like 'Sarah is home, 3:47pm' notifications without triggering alarms.
Is the AI processing happening in the cloud or on my device?+
Both. Initial classification runs on-device for the fastest response (under 200ms). For edge cases or unusual scenes, the system falls back to cloud-based processing for higher accuracy. Either way, your video is encrypted in transit and at rest.