Home / AI Edge Computing / AI Video Analytics Server

AI Video Analytics Server — 96, 128 or 352 Channels of Centralised AI Analysis

Q: How do I choose between 96, 128 and 352 channels?

Choose by the number of camera streams requiring simultaneous analysis, not the total camera estate. QG-6181U96 covers up to 96 channels, QG-6182U128 handles 128 channels, and QG-6184U352 supports up to 352 channels.

Q: What is large-model re-checking and why does it matter?

Large-model re-checking re-evaluates alarm events detected by the fast small-model set with a larger model before publication, suppressing false alarms in outdoor and industrial scenes.

Q: Does the server support GB28181 platform cascade?

Yes. GB28181 platform cascade is supported for integration with higher-level management platforms.

Q: Can the server manage edge AI boxes at remote sites?

Yes. The central management platform provides edge AI node management including remote status monitoring, remote firmware and algorithm deployment, and unified alarm aggregation.

For programmes where the camera count runs into the hundreds, ADHOOPU AI compute servers centralise the analytics workload: 96, 128 or 352 channels of 1080P AI analysis in 1U, 2U or 4U, with optional large-model re-checking and centralised algorithm management.

AI VIDEO ANALYTICS SERVER96 / 128 / 352 CHANNEL CENTRALISED AI ANALYTICSMULTI-CHANNEL AILARGE-MODEL RE-CHECK

AI VIDEO ANALYTICS SERVER · AI VIDEO ANALYSIS SERVER · AI COMPUTING SERVER · MULTI-CHANNEL AI VIDEO ANALYTICS SERVER · 128 CHANNEL AI VIDEO ANALYTICS · CENTRALIZED AI VIDEO ANALYTICS · DISTRIBUTED AI VIDEO ANALYTICS

Three Models, One Architecture

All three servers share the same software stack and algorithm library. The difference is channel capacity and chassis size — choose by camera count and rack availability.

Model	Chassis	Max Channels	Best For
QG-6181U96	1U rack-mount	96 channels of 1080P	Single-site programmes, rack-constrained rooms
QG-6182U128	2U rack-mount	128 channels of 1080P	Multi-building campuses, industrial parks
QG-6184U352	4U rack-mount	352 channels of 1080P	Large programmes, city and enterprise scale

Each chassis contains 2 to 22 edge compute units plus an industrial control mainboard. Channel capacity is a function of the number of compute units installed.

Specifications

Verified platform parameters of the ADHOOPU AI compute server line.

Product Type	Centralised AI video analytics server
Models	QG-6181U96 (1U) / QG-6182U128 (2U) / QG-6184U352 (4U)
Max Channels	96 / 128 / 352 channels of 1080P
Compute Units	2 to 22 edge compute server units + 1 industrial control mainboard
Processor	Domestic chipset (default: Intel industrial control mainboard)
Operating System	Linux
Video Input	ONVIF / RTSP / GB28181 / M51C / WEBRTC / RTMP
Decode	H.264 / H.265
Resolution Support	Unrestricted — 720x480 up to 3840x2160
Algorithms	46+ small-model algorithms, large-model re-check supported
Face Database	Up to 300,000
Network	2 x 100M/1000M adaptive Ethernet
Protocols	TCP / UDP / HTTP / HTTPS / MULTICAST / DHCP / FTP / NTP / RTSP / GAT1400 / MQTT
Platform Cascade	GB28181 platform cascade supported
Storage	Local storage with centralised synchronised storage; 20s pre/post-event video
Management	Central management platform: dashboard, alarm management, device management
Network Architecture	Internal/external network isolation; compute units on LAN, separate port for external network
Warranty	3 years

Colour-coded accuracy

Engineering-verified detection and recognition rates.

Recognition Accuracy

Face capture ≥99% — misidentification <0.5%
Whitelist face recognition >99.5%
Face database up to 300,000
Human body capture ≥95%, false capture <1%
Non-motor vehicle capture ≥95%, false capture <1%
License plate capture ≥95%, recognition ≥95%
Motor vehicle capture ≥90%, false capture <1%
Other alarm categories ≥90%

Platform Capabilities

Data dashboard: algorithm count, device count, stream count, alarm trends
Alarm centre with classification, search and export
Algorithm configuration, orchestration and one-click deployment
Offline video upload for algorithm validation
False-alarm feedback into incremental model training
Visual algorithm test interface — upload any image, run any model
Image search and face database search

OEM / ODM

Video analytics server OEM under your brand
Custom algorithm set and model training
Custom platform branding and API integration
See the OEM / ODM programme

Edge, Centre, or Both

The AI server line is designed to work with edge AI boxes, not instead of them. Most large programmes use both layers.

Centralised Only

All camera streams are sent to the server for analysis. Simplest architecture, best where cameras are concentrated in one or a few locations with adequate network capacity.

Edge Only

An edge AI box at each site handles analysis locally and only events travel the network. Best for many distributed small sites with limited bandwidth.

Edge + Centre (Recommended at Scale)

Edge boxes handle real-time detection at source; the central server aggregates, manages, performs large-model re-checking and provides the unified platform view.

AI Server FAQ

How do I choose between 96, 128 and 352 channels?

Choose by the number of camera streams you need to analyse simultaneously, not the total camera estate. If only a subset of cameras require analytics, size to that subset. QG-6181U96 covers single-site programmes up to 96 channels, QG-6182U128 handles 128 channels for multi-building campuses, and QG-6184U352 supports up to 352 channels for large-scale programmes.

What is large-model re-checking and why does it matter?

Large-model re-checking means alarm events detected by the fast small-model set are re-evaluated by a larger model before being published. This suppresses the false alarms that make conventional detection unusable in outdoor and industrial scenes — the difference between an operator trusting the system and ignoring it.

Does the server support GB28181 platform cascade?

Yes. GB28181 platform cascade is supported, allowing the server to integrate with higher-level management platforms — important for government, public safety and multi-level enterprise deployments.

Can the server manage edge AI boxes at remote sites?

Yes. The central management platform provides edge AI node management, including remote status monitoring, remote firmware and algorithm deployment, and unified alarm aggregation. This is the standard architecture for multi-site industrial and enterprise programmes.

How does the platform handle false alarms over time?

False alarms can be pushed back into the system as training data for incremental model improvement, and confidence thresholds are tunable per algorithm and per channel. Over a project life cycle this means accuracy improves rather than degrading.

Is the AI server available for OEM / private label?

Yes. It is available as video analytics server OEM hardware under your brand, with custom algorithm sets, custom platform branding and API integration. See the OEM / ODM programme.

ADHOOPU

Professional AV manufacturer — Integrated AV Platforms, Distributed AV over IP, and AI Edge Computing & Video Analytics. OEM/ODM for system integrators, distributors and AV brands worldwide.

19 years of AV engineering · Shenzhen, China · 3-year warranty

Products

Integrated AV Platform QG-1169 AV over IP Encoders ST52K Encoder SR52K Decoder AI Edge Computing 615 AI Analytics Box

Solutions

Smart Courtroom Surgical Teaching Command Center Education Meeting Room Smart Security

Company

OEM / ODM About ADHOOPU FAQ Resources Contact

HMW Technology Holdings (Shenzhen) Co., Ltd. · Room 1206, Block A, Dachanwan, Jingang Commercial Building, Xixiang Street, Bao'an District, Shenzhen, China
WhatsApp +86 138 2352 5531 · Email jimmy@adhoopu.com · Web https://www.adhoopu.com
© 2026 ADHOOPU. Integrated AV Platform · AV over IP Encoder / Decoder · AI Video Analytics · Professional AV OEM Manufacturer.