Crowd-grade recall
Detects individuals even when partially occluded in dense queues at turnstiles, gates, and terminal walkways.
Innomium Vision · Edge YOLO
Person detection
built for crowds.
Sentinel is Innomium's cutting-edge YOLO model — very light, very strong, and tuned for the hardest real-world scenes. From airport concourses to stadium gates, it finds every person with 92% accuracy.
- Accuracy
- 92%
- Model size
- 19 MB
- Runtime
- Edge / Browser
Live detection
High-density crowd · Stadium entrance
Proven performance
From raw scene to precise detections.
Sentinel maintains high recall in dense, overlapping crowds — the kind of scenes where generic detectors miss people or flood you with false positives.
Confidence you can trust
Scoring-aware post-processing and TTA consensus keep high-confidence detections clean in complex scenes.
Deployment-ready
A compact 19 MB ONNX model suitable for edge cameras, kiosks, and in-browser demos without a GPU farm.
Capabilities
Enterprise vision, without enterprise weight.
Adaptive confidence
Dynamic thresholds respond to scene density — maximizing recall when few people are present and precision when the crowd thickens.
Ultra-light ONNX
~19 MB footprint loads fast and runs on CPU via ONNX Runtime — in the cloud, on edge hardware, or directly in the browser.
TTA consensus
Test-time augmentation cross-validates detections across flipped views to reduce false alarms before they reach your pipeline.
Privacy-first
This demo runs inference entirely in your browser. Frames never leave your device — ideal for regulated environments.
Use cases
Where Sentinel delivers.
Airports & transit
Passenger flow, queue monitoring, and gate-level occupancy in high-throughput terminals.
Stadiums & events
Entry-point crowd analysis, congestion detection, and real-time headcount at scale.
Stores & malls
Foot traffic analytics and zone density with a model small enough for in-store edge boxes.
Public spaces
Situational awareness for campuses, plazas, and smart-city pilot deployments.
Interactive demo
Upload your own scene.
Run the production ONNX model in your browser — same pipeline, TTA and adaptive confidence included.
No image loaded
Awaiting detection
Detection JSON
[]
Deploy Sentinel in your vision stack.
Very light. Very strong. 92% accuracy in crowds. Built by Innomium for airports, venues, and edge deployments that cannot afford to miss a person.