Release Notes - CVEDIA-RT - 2026.1.0 ¶

Distributed Deployment for NX Witness and Milestone XProtect Plugins ¶

In this release, we are excited to introduce Distributed Deployment support for both the CVEDIA-RT Plugin for NX Witness and the CVEDIA-RT Plugin for Milestone XProtect. This allows the plugins to connect to one or more remote CVEDIA-RT Servers for AI processing, enabling more flexible and scalable deployments.

The plugins are designed to automatically discover and connect to available CVEDIA-RT Servers within the same network. This feature simplifies the setup process and ensures that AI processing can be distributed across multiple servers as needed.

Features ¶

VMS plugins automatically discover and connect to multiple remote CVEDIA-RT servers for AI processing.
The plugin will automatically attempt to reestablish the connection to a different box under the same namespace if the current one becomes unreachable, or if the license on the current box does not support the required features or number of streams.
RT Servers can be grouped in different namespaces. This allows organizing multiple RT Servers based on different licensed features or different departments within the same network.
In the Expert Settings of the plugins, it is now possible to configure a specific RT Server box to connect to, overriding the automatic selection mechanism. This is useful in scenarios where the RT Server are not discoverable via the network.
The license panel now allows the selection of the desired RT Server to manage licenses for.

Migration Notes ¶

The new Distributed Deployment requires the installation of the CVEDIA-RT Server separately on one or more machines in the network. The previous full Server+Client installer is no longer supported. Follow the instructions below based on your desired deployment scenario.

Keep the Same Machine Deployment ¶

If you want to keep the CVEDIA-RT Server on the same machine as the VMS server, run the new lightweight installer for the CVEDIA-RT Plugin for NX Witness or Milestone XProtect on the respective VMS server.

When prompted, select the option to install the CVEDIA-RT Server on the same machine.

When asked, keep 127.0.0.1 as the server address.

Info

The installed licenses and the existing configurations will be preserved during the upgrade.

Distributed Deployment on Different Machines ¶

If you want to move the CVEDIA-RT Server to a different machine, you can follow these steps:

Back up and revoke any existing licenses on the device with the previous version of the plugin before proceeding with the upgrade.
Install the CVEDIA-RT Server on one or more machines in your network.

When asked, select 0.0.0.0 as the server address.

You don't need to activate the licenses yet. It will be done later from the plugin license panel.
Run the new lightweight installer for the CVEDIA-RT Plugin for NX Witness or Milestone XProtect on the respective VMS server. When prompted, select the option to skip the installation of the CVEDIA-RT Server on the same machine.
After installation, use the plugin license panel to manage the licenses on the new CVEDIA-RT Server(s) in your network.

Introducing the AI Appliance ¶

The AI Appliance is a Linux-based stand-alone box with AI acceleration (CPU/GPU/NPU) that comes with Cockpit admin panel and the CVEDIA-RT Server application pre-installed. It allows adding analytics capabilities without modifying or upgrading existing servers.

Key functions:

Offers an easy-to-use administration panel for the device
Runs the RT Server (AI inference engine)
Allows a fast deployment of multiple RT Servers in the network
Exposes analytics output to any VMS or 3^rd-party system via plugins or REST API
Simplifies the deployment without requiring upgrades to existing servers.

The AI Appliance can be purchased as a pre-configured hardware device or manually deployed on existing hardware that meets the minimum requirements.

For more information, please refer to the AI Appliance documentation.

CVEDIA-RT Analytics ¶

Analytics Improvements ¶

Greatly improved the Crowd Estimation model by reducing false positives in multiple scenarios.
Improved Person Vehicle Animal Detection model
- improved detection of vehicles in thermal
- reduced number of FPs on animals (mostly birds)
- improved detection of small/blurred people on low quality footage
- improved detection of people, in particalar bikers
Added animal classification: bear, elephant, wolf, cat, dog, bird, deer.
Changed configuration type for "Face covered" feature from "zone" to "virtual line", allowing detection to trigger only when people are moving in a specific direction. Also removed the class selection from REST API, as it now only applies to persons.
Improved the tracker performance and accuracy.
Improved attribute capture logic for objects to ensure we keep track of "state" attributes such as fallen, armed, surrender, etc. throughout the entire object's lifecycle.
Reduced false positive outputs for surrender, assisted, phone, glasses, and smoking attributes.

Bug Fixes ¶

Fixed problem with triggering multiple queued up events when using fallen_person area
Fixed full-screen exclusion area not working correctly
Fixed a problem where using vehicle attribute restrictions on a zone would prevent it from generating person events

Plugin for NX Witness ¶

New Features ¶

Implemented a new lightweight installer for the CVEDIA-RT Plugin for NX Witness, Piko, Wave, etc., which installs only a lightweight connector on the Media Server. It's expected to be paired with one or more CVEDIA-RT Servers running in the same network. Please refer to the Migration Notes section above for more details.
The enhanced search option has been simplified and now only requires enabling the "Enhanced Search" checkbox. It will automatically classify people, vehicles, and animals.
The enhanced search now also extracts animal species: bear, elephant, wolf, cat, dog, bird, deer.
The Linux installer now asks to accept the EULA before proceeding with the installation. For automated installations, refer to the NX Witness Plugin Installation documentation and select the "Linux" tab.

Bug Fixes ¶

Fixed an issue where a single crowding event could create multiple redundant bookmarks, ensuring only one bookmark is generated per crowding event.

Plugin for Milestone XProtect ¶

With the introduction of the new AI Appliance support, the CVEDIA-RT Plugin for Milestone XProtect can now connect to multiple remote CVEDIA-RT servers for AI processing, also on Linux devices.

New Features ¶

Implemented a new lightweight installer for the CVEDIA-RT Plugin for Milestone XProtect, which installs only the CVEDIA-RT Connector Windows service on the Milestone Management Server. It's expected to be paired with one or more CVEDIA-RT Servers running in the same network.
The enhanced search option has been simplified and now only requires enabling the "Enhanced Search" checkbox. It will automatically classify people, vehicles, and animals.
The enhanced search now also extracts animal species: bear, elephant, wolf, cat, dog, bird, deer.

Please refer to the Migration Notes section above for more details.

Improvements ¶

Added date and time to the smart search result thumbnails
Showing warnings for invalid values in analytics settings
Added an option in the search to match any or all of a person's "other attributes" like glasses, hat, phone, etc.
Reduced the size of analytics events stored in the XProtect SQL database

Bug Fixes ¶

Fixed instance creation failure for cameras with non-ASCII names (e.g., names with accented characters)
Fixed high memory usage in analytics setup for high-resolution cameras
Fixed a crash in the Connector service when a recording server contains devices with corrupted metadata
Improved Connector Service stability and reconnection handling after temporary loss of connection to Milestone XProtect Management Server
Resolved an issue where the XProtect Connector service could occasionally get stuck during manual shutdown.
Fixed instance configurations sometimes being lost during connection interruptions
Fixed an issue that was preventing the rendering of the Overlay Statistics panel when the camera name contains special chars

CVEDIA-RT Engine ¶

Windows Installer Changes ¶

Separate installers: CVEDIA-RT is now distributed as two separate installers - RT Server and RT Studio
RT Server (required): Core runtime service for production deployments
RT Studio (optional): Desktop application with UI and demo solutions for development and testing

Info

RT Server must be installed first
RT Studio is an optional add-on that requires RT Server
Versions must match exactly

The installer for the RT Server now allows you to configure the binding IP of the web server for the REST API (default 127.0.0.1). This setting can also be changed after the installation using the rtcmd config command-line tool. For more information, refer to the RT Server Configuration documentation.
Additional command-line options have been added for silent installation. Refer to the updated section in the Running RT Server on Windows documentation for more details.

Linux Installer Changes ¶

New command-line for the Linux installer:

curl -sqko - http://get.cvedia.com/rt-server | sudo bash

The installer asks to accept the EULA before proceeding with the installation.
The installer now allows you to configure the server binding IP of the web server address for the REST API (default 127.0.0.1). This setting can also be changed after the installation using the rtcmd config command-line tool. For more information, refer to the RT Server Configuration documentation.
To pre-accept the EULA or override the server binding IP, refer to the updated environment variables in the Running RT Server on Linux documentation.
the installer now configures the cvedia-rt service to run under the cvedia user (automatically adding it to the video and render groups for hardware acceleration), and migrates any existing NX licenses to /home/cvedia/.config/cvedia/licenses with correct ownership.

RT Command-Line Tool Improvements ¶

Added ability to change the configuration of the Web Server using the rtcmd config command-line tool.

For more information, refer to the RT Server Configuration documentation.

REST API ¶

⚠️ Breaking Change: the /v1/securt/instance/{instanceId}/attributes_extraction REST endpoint has been changed to accept a classes array (["Person", "Animal", "Vehicle"]) instead of the previous mode field (Off, Both, Person, Vehicle).

Web Panel ¶

Added "animal" to the list of classes for attribute extraction configuration.

Licensing Improvements ¶

Improved license key management by allowing removal of unactivated licenses without requiring force option
Improved logging for license operations

Drivers and Platforms ¶

Added DEEPX inference support for x86 and aarch64 platforms
Added Jetson hardware accelerated decoder
Added support Intel acceleration on Linux

Bug Fixes ¶

Fixed GStreamer shutdown sequence that could prevent RT from shutting down while an RTSP stream is unresponsive
Fixed RT Server Linux service not starting in certain configurations

Modelforge ¶

Implemented DEEPX model conversion support
Added model URI alias system for easier deployment and hardware-specific model mapping