Custom models
Your critical event, learned from your footage.
Your own event? We assess feasibility and train a model adapted to your footage.
Approach
Six steps. A model that knows your environment.
- Step 01
Defining the event
We turn what you want to detect into signals a camera can see: who, what posture, where, for how long.
- Step 02
Your footage, lookalikes included
We pick the clips where the event happens, and the ones where it only looks like it: someone crouching, bending over, an object on the floor.
- Step 03
Pre-annotation, human review
A foundation model proposes. Our annotators correct every detail, then validate.
- Step 04
Training on the hard cases
The model learns what the event is as much as what it isn’t, to keep false alerts down.
- Step 05
Tracking over time, and rules
One frame isn’t enough. The person is followed from frame to frame, and the alert fires only if they stay down, in the zone, during the hours you choose.
- Step 06
Tuned for your site, improved by your feedback
The model is optimized to run on your box or server. Every alert you confirm or dismiss feeds the next version.
FAQ
Common questions. Clear answers.
Demo
Let’s see what matters in your environment.
Tell us about your sites. We’ll show you Aquilix on your case.