Skip to content

Release Notes - CVEDIA-RT - 2024.2.1

As usual, every new release brings updated models, new features and improvements, and bug fixes.

Please check below in detail what's new.

CVEDIA-RT Plugin 3.0 for NX Witness and White Label Variants

  • Added the Object Guarding analytics.
  • Added the Object Left Behind analytics.
  • Added the Tailgating analytics.
  • Added the ability to change the performance mode in SecuRT to either Performance or Balanced. This is a global setting that applies to all the AI-enabled cameras.
    • The Performance mode is the default suggested profile, offering optimal performance while maintaining high precision for common security analytics. This mode uses the secondary camera stream, which should be set to a resolution of 480p at 10 FPS.
    • The Balanced mode utilizes a larger AI model and the primary camera stream, enhancing the detection of smaller targets at a distance or in more crowded environments. Enable this mode only if the Performance mode proves insufficient.

  • Automatic stream selection can be disabled when switching between Performance and Balanced mode. This allows you to force the use of a specific stream per camera.

    Example Scenarios for Overriding:

    • Use Case 1: Detecting smaller targets without significantly impacting the number of AI streams:

      If you need to use the primary stream for detecting smaller targets but a larger AI model limits the number of available streams for AI processing, follow these steps:

      • Select Performance mode.
      • Disable Automatic Stream Selection.
      • For each camera, select the primary stream from the Camera Stream dropdown in the main panel of the plugin’s camera settings.

    • Use Case 2: Requiring a larger AI model without relying on an excessively demanding primary stream:

      If you require a larger AI model but the primary stream cannot be set to a lower FPS, follow these steps:

      • Select Balanced mode.
      • Force the use of the secondary stream, which can be configured to a higher resolution.

    These adjustments allow for more flexible and efficient use of system resources based on specific needs and constraints.

  • Added the ability to switch between RT Smart Detection based on motion analysis and Fullframe detection.

  • Statistics now show used decoder, codec, and resolution.
  • Event thumbnails have anti-aliasing.
  • Removed debug information from object thumbnail details.
  • Camera orientation can now be set in the plugin (it should match the one set in the camera settings).
  • Running cameras with unsupported codecs show as unsupported in the statistics panel.

Bug Fixes:

  • Instances do not start after adding a valid license.
  • Fixed an issue that was preventing the deactivation of a license using the UI in the Plugin for NX Witness.
  • In some cases, motion masks were invalid, causing missing events.
  • Fixed Balanced mode never using hardware acceleration for decoding.
  • Fixed random buffer issues in Balanced mode.

CVEDIA-RT Plugin 2.0 for Milestone

  • Added Appearance Search Attributes for vehicle types and clothing and vehicle colors.
  • Added more information to the statistics panel.

SecuRT Solution 12.1

  • Added support for multiple performance modalities: Performance, Balanced, and Accurate, respectively using larger AI detectors for Person, Vehicle, Face, and License Plate classes.
  • Added the ability to change the performance mode in SecuRT to either Performance, Balanced, or Accurate.
  • Changed movement direction sensitivity, improving results in event detection.
  • Loitering use-cases have been improved when using motion-guided inference.

RT Studio

  • Added the ability to configure Crowd Estimation zones in SecuRT.
  • Added the ability to configure the performance mode in SecuRT.

Bug Fixes:

  • Fixed an issue that was preventing the Event detail popup from opening when more than one event is listed.
  • Fixed an issue that was preventing the export of RTMP streams.
  • Fixed an issue with the save button when an instance is read-only.

CVEDIA-RT

  • Active decoder being added to the system registry.
  • Statistic events now include the resolution.
  • Added performance profile concept (different model sizes based on the performance profile). The default profile is set in the rtconfig.json configuration file.
  • Core API now supports input orientation.
  • Added C API to set autostart.
  • Updated the default ip_address setting of the web server to 127.0.0.1 (localhost) from 0.0.0.0 to enhance security by restricting access to the local machine only.

Bug Fixes:

  • Fixed an issue that was causing the detector to return too many detections.
  • Fixed concurrency issues with TensorRT.
  • Fixed concurrency NVIDIA decoder crash.
  • Added more verbose decoder logging.
  • Fixed motion mask recalculation.
  • Fixed autostart and auto-restart of instances.
  • Fixed an issue that was preventing the export of RTMP streams.
  • Fixed an issue in PowerShell that was causing an error after running rtcmd.exe.

Windows Installer

  • Fixed potential failure to download vc_redist packages.

Known issues

  • CVEDIA-RT plugin is not compatible with Gen 6 (v6.x.x) releases of Nx Witness, Wisenet WAVE, and Piko.
  • Nvidia decoding has poor performance. This will be improved in the next release.
  • Balanced mode puts a heavy load on the CPU when running at high frame rate (above 10FPS). This will be improved in the next release.