Edge Computing

Edge computing processes data on or near the device that generated it — for video analytics, that means running AI inference on the camera itself or on a local server rather than sending streams to the cloud.

How It Works

In an edge video analytics deployment:

1. Cameras (or a nearby edge server) run trained AI models locally on specialized hardware — GPUs, NPUs, or VPUs.
2. Only metadata — detections, alerts, snapshots — is sent over the network.
3. Raw video stays on-site unless a specific incident requires escalation.

This architecture trades some flexibility (harder to retrain) for major gains in latency, bandwidth, privacy, and resilience.

Why It Matters

Cloud-only video AI has three structural problems that edge solves:

• Latency. Streaming 1080p H.265 to the cloud, running inference, and returning an alert adds hundreds of milliseconds to seconds. Edge inference runs in under 50 ms.
• Bandwidth. A single 4 Mbps camera uses ~43 GB/day. 500 cameras uploaded 24/7 is not cost-effective.
• Privacy and sovereignty. Many jurisdictions prohibit shipping biometric video outside national borders.

IncoreSoft's VEZHA platform supports edge, cloud, and hybrid deployments, letting each site choose the right balance.

Use Cases

• Time-critical alerts — gun detection, fall detection, smoke/fire where seconds matter
• Bandwidth-constrained sites — remote industrial, maritime, or mining locations
• GDPR/HIPAA environments — biometric or medical footage must stay on-premise
• Resilient deployments — analytics must keep working during network outages
• Retail chains — hundreds of stores without per-store cloud cost

Frequently Asked Questions

What's the difference between edge and cloud video analytics?

Cloud runs AI in centralized data centers — flexible and easy to update, but higher latency and bandwidth cost. Edge runs AI on-site — faster and private, but requires capable local hardware.

Can edge analytics run on existing cameras?

Some modern cameras ship with on-board AI chips (e.g., Axis ARTPEC-8). For older cameras, IncoreSoft deploys an edge server or appliance that connects over RTSP and runs analytics locally.

How do edge deployments handle model updates?

A hybrid architecture is common: models are trained in the cloud and pushed to edge devices over secure channels on a scheduled basis, combining cloud flexibility with edge performance.