The AI helps filter out false alarms and makes it easier to find specific footage. It's not perfect—you'll still get occasional false positives—but it's significantly better than basic motion detection.
AI processing runs on the AI Edge Analytic Box locally; optional cloud sync for search and multi-site dashboards. We don't use customer video to train models. SOC 2 Type II reports are available on request.
RKNN models run on the AI Edge Analytic Box NPU — safety, counting, traffic, and access control without waiting on the cloud. Add optional cloud sync for search and multi-site dashboards. Explore our four application suites.
Six deep learning modules that transform surveillance cameras into intelligent sensors
The AI can tell the difference between your mail carrier and a stray cat. No more 3am alerts because a moth flew past the camera.
Type 'person in red jacket' or 'blue delivery truck' and jump right to it. No more scrubbing through 8 hours of footage to find a 10-second clip.
Our AI filters out rain, shadows, and tree branches. You get notified when a person shows up—not every time the wind blows.
Heat maps show you which aisles get traffic, where people linger, and why one entrance is always crowded. Great for retail layouts and event planning.
Automatically captures plate numbers as vehicles pass by. Set up a watchlist and get alerted when specific plates arrive—useful for deliveries, employees, or unwelcome visitors.
Real-time headcounts for stores, venues, or anywhere you need to track occupancy. Great for capacity limits, shift planning, or just knowing how busy things get.
Deep learning architecture, inference performance, and integration details
Real-world accuracy depends on camera angle, lighting, and which model you enable. A camera pointed into the sun will do worse than one with decent backlight compensation. You will still see occasional false positives from shadows or wildlife. The useful change is fewer noise alerts than basic motion, not a perfect score.
Edge AI runs on the AI Edge Analytic Box at your site—construction safety, people counting, traffic, LPR, and facial recognition with 15–30ms typical latency. It works when internet drops. Cloud AI adds semantic video search, cross-site dashboards, and long-term trend reports. Time-critical alerts use edge; search and multi-site ops use optional cloud sync.
Object detection uses compact models on the edge NPU (RKNN). Heavier models can run in the optional cloud path when you want higher accuracy. Feature extraction uses a re-identification network. Semantic search uses CLIP embeddings in a vector database. License plates use OCR with regional models. We update models as better architectures emerge.
No, and we're contractually bound to that. Our models train on licensed datasets (COCO, ImageNet, proprietary collections we've licensed). Customer video never leaves your data silo except for the specific features you enable. We can provide DPAs and data handling documentation for enterprise procurement.
Yes—this is what makes cloud AI worthwhile. Type 'red truck' or 'person carrying box' and we'll find matching clips in seconds. Under the hood, we embed video frames using CLIP, store vectors in a similarity search index, and match against your query. It searches across all cameras and can process months of footage. Way faster than scrubbing manually.
VMS platforms like Milestone and Genetec offer analytics plugins, but they're often separate add-ons with their own licensing. We bundle analytics into the cloud subscription—no extra servers to deploy or plugins to manage. If you already have XProtect or Security Center, we can feed our AI events into those systems via API.
We're GDPR-aligned with EU data residency options. Face detection (presence) runs on all tiers. Full facial recognition (watchlist matching, 1:N) requires an Enterprise on-prem deployment plus documented lawful basis, consent/signage, and tenant-isolated galleries—we provide DPIA templates and regional enablement controls. Customer video is never used to train shared models.
Depends on your cameras. If they output clean IR images, detection accuracy stays high. If they're using cheap sensors that get grainy at night, accuracy drops. We recommend cameras with true WDR (120dB+) and Starlight sensors for 24/7 analytics. The AI can only work with what the camera gives it—garbage in, garbage out.
We'll show you the AI in action—with your actual footage. No commitment, just a demo so you can see if it's a good fit.
Want to see the full platform first? Check out all features