CORE

NQVCA

On-device AI video analytics for NQNVR

AI analytics that never leaves the device

The AI analytics subsystem built into NQNVR. Identifies people, vehicles, faces, fire and license plates in real time without any cloud, and raises events from intrusion, counting, fall and wrong-way rules.

On-premise, no cloudOpenVINO · Hailo-8 · RKNNKorean license plate recognitionMulti-object tracking

Platforms

Ubuntu 24.04 LTSWindows 10 / 11RK3588 (ARM)

AI detectors

CPU · NPU

Inference backend auto-selected

Rule types

Live

Rules apply instantly

Sales inquiry All products

Overview

NQVCA consists of an analytics engine (a background service) that receives video from the NQNVR recording engine, decodes it and runs AI object detection, and a viewer (NQVcaViewer) that overlays detections and configures rules. Three inference backends, Intel CPU (OpenVINO), Hailo-8 NPU and RK3588 NPU (RKNN), sit behind one interface: an NPU is used automatically when present, otherwise the CPU takes over. Detections are stored as best shots and events in the NVR and forwarded over HTTP and MQTT to external systems or applications such as NQVehicleTracker.

Notes

NQVCA is an optional component of NQNVR. The [VCA] button appears in the management console only when the VCA package is installed.

Component app

NQPlateExtractor

Component app

License plate extraction from recordings

Windows 10 / 11

Learn more

NQVehicleTracker

Component app

Vehicle location and tracking

Windows 10 / 11Ubuntu 24.04 LTS

Learn more

Key Features

Key features

Six detection engines

Person, vehicle, face, fire, license plate (Korean LPR) and face re-identification (embedding-based 'find the same person'). Enable only the engines each channel needs.

Rule-based events

Detection, polygon intrusion, uni/bi-directional line counting, plate recognition ROI, fall and wrong-way. Rule changes apply immediately without a restart.

Automatic CPU/NPU selection

Finds the model that matches the installed accelerator and picks Hailo-8, RK3588 NPU or OpenVINO automatically. Configuration stays the same when hardware changes.

Korean plate recognition

A dedicated Korean plate model with format validation reduces misreads, and repeated reads of the same plate are suppressed.

Best shots

Stores and searches the best frame of every object and jumps straight to the recording at that moment. Retention is configurable.

Thermal camera integration

Reads temperature data from thermal cameras to raise fire and temperature-alarm events.

HTTP and MQTT outputs

Publishes events over HTTP POST and MQTT to external systems and apps. NQVehicleTracker and NQNMS subscribe to the same stream.

Built-in benchmark

Measure per-model processing speed in the viewer to confirm headroom on site hardware.

Screens

Specifications

Analytics
Detectors	Person · vehicle · face · fire · license plate · face re-ID
Rules	Detection · intrusion · counting · plate · fall · wrong-way
Tracking	Multi-object tracking with track-ID overlay
Sensitivity and rate	Three sensitivity levels and detection rate per channel

Platform
Inference	OpenVINO (Intel CPU) · Hailo-8 NPU (M.2/PCIe) · RKNN (RK3588)
Face model	Embedding-based face re-identification, commercially licensed model
Input	Direct feed from the NQNVR recording engine (no re-encoding), up to 64 channels
Outputs	NVR event store · best shots · HTTP POST · MQTT
Component apps	Analytics engine (background service) · NQVcaViewer · NQPlateExtractor · NQVehicleTracker
Licensing	VCA / LPR feature flags in the NQNVR license