Skip to main content
Version: 2.9.1 (latest)

Release notes

2.9.1 (13 Dec 2024)

What's New

  • The event specification has been expanded for deeper integrations with external systems: ROI (Region of Interest) and line crossing events now include data on UUID, name, and points of the ROI/line.
  • Introduced a threshold check for the confidence of keypoint detection in skeleton tracking. This will allow fine-tuning the detection accuracy for specific business tasks.

Improvements

  • Updated skeleton display in preview:
    • Keypoints with low detection confidence are no longer displayed;
    • If at least one keypoint forming a bone has low detection confidence, the bone is displayed as a dashed line.
  • Updated Face SDK version to 3.22.2.

Bug Fixes

  • Fixed incorrect triggering of face hide control events when crossing a line in the opposite direction.
  • Now only finalized line crossings are processed.
  • It is now possible to process the video file in a way that events are sent to the server with the date corresponding to the video file's recording time, rather than the current date.

2.9.0 (12 Sep 2024)

What's New

  • Added support for using regions of interest and intersection lines for detecting faces and body parts. Face or body part tracking within a marked area is done by the center of detection, while body tracking is based on the center of the lower boundary of the bbox.
  • Added the ability to receive raw detection data from the face and body pipeline via the MQTT event queue.
  • Added the option to lock configuration changes from the web interface.
  • Added the ability to select image formats sent to OMNI Platform (.png, .jpeg, etc.).

Bug Fixes

  • Fixed a bug with USB camera order changing in the camera list after restarting OMNI Agent.
  • Fixed inability to add more than one USB camera from the same manufacturer.
  • Fixed rare issue of “template version confusion” at startup, which temporarily prevented face identification.

2.8.1 (07 Aug 2024)

Bug Fixes

  • Optimized profile reading: Fixed an issue that caused slow reading of profiles from the file system during operations with the profile database.
  • Efficient profile database update: Now when updating the profile database in OMNI Platform, only the changes (diff), not the entire database, are updated in the OMNI Agent profile database.

2.8.0 (24 May 2024)

What's New

  • Added facial concealment detection (BETA). Now, OMNI Agent can identify instances where individuals attempt to hide their faces (by turning away or using scarves, hats, glasses, or masks) while passing through controlled areas. The new detector enables swift responses to evasion attempts from facial recognition systems and aids in compiling statistics on such disciplinary breaches in workforce time management systems.
  • Added the ability to send events via webhooks directly from OMNI Agent. No longer limited to OMNI Platform database, this enhancement offers flexibility for seamless integration with client infrastructures.
  • Added the ability to enable / disable specific HAR events such as falls, fights, sitting and lying, and tailor the confident detection time for each event. Customize your tracking preferences and define the duration threshold for significant events, ensuring only precise and relevant alerts for your operational needs.

Bug Fixes

  • Removed the ability to enable event sending for disabled pipelines in OMNI Agent. For example, previously, when the HAR toggle was disabled in Settings, you could still enable the "Send HAR events" toggle. Now, this is no longer possible.
  • Fixed rare connection errors of OMNI Agent to OMNI Platform via secure connection (HTTPS). Very rarely, the Agent could not connect to the Platform because it could not find the Platform's certificate in its list of trusted certificates.

2.7.1 (16 Apr 2024)

Bug Fixes

  • Fixed a bug where Hikvision emulation settings would reset upon reactivating OMNI Agent on OMNI Platform, unless OMNI Agent was restarted after modifying them.

2.7.0 (27 Mar 2024)

What's New

  • Updated HAR model for enhanced human action recognition accuracy.
  • Introducing the feature to process video files directly from the web interface.
  • New option to save storage space in OMNI Platform by disabling the transmission of original frames in events.
  • Now object coordinates from the original frame are included in events, enabling seamless person movement tracking without transmitting original frames.
  • Sending facial data in Hikvision smart camera format from OMNI Agent, ensuring smooth integration of standard and "smart" cameras for streamlined image processing.
  • Optimized template generation setting via the web interface. You can now work with the initial biometric template (built from the first best shot of the face) or initiate a search for the best shots across the entire track to generate higher-quality templates.
  • Enhanced face identification accuracy by continuously searching for the best face shots within the track, boosting comparison precision.

Fixes

  • Fixed intermittent loss of generated template data or identification data.

2.6.0 (19 Feb 2024)

What's New

Bug Fixes

  • Fixed a bug with incorrectly functioning face reidentification, where upon reentry into the camera's field of view, the person got a new ID.

2.5.1 (22 Feb 2024)

Bug Fixes

  • Fixed disabling of the face detection pipeline when changing desired_score in the web configurator.

2.5.0 (13 Dec 2023)

What's New

  • HAR (Human Activity Recognition) and ROI (Region of Interest) detections are now transmitted to OMNI Platform as event-service events. This provides a unified integration approach for data transmission to external systems for both face identification/non-identification and action recognition.
  • It's now possible to eliminate duplicate identification and human action recognition events caused by changes in perspectives, head rotations, people crossing tracks, or brief obstructions (event cooldown).
  • Added the ability to limit the camera's field of view through configuration files and the dashboard. This helps exclude duplicate detections/identifications from cameras whose fields of view intersect and overlap, as well as to eliminate false detections/identifications at the edges of the lens where optical distortion occurs.
  • Added full support for body tracking on ARM64 (Jetson), along with MQTT protocol support for output data and the ability to configure process transmission via MQTT. This makes it easier to integrate OMNI Agent with end devices in solutions where OMNI Platform server is not required.
  • It's now possible to change videoworker settings on OMNI Agent through configuration files to ensure optimal performance for the required quality of face detection and identification.

Bug Fixes

  • Previews are now only rendered when viewing a camera in the web configurator, saving processor resources.
  • Improved body tracking and HAR (Human Activity Recognition) detection quality.
  • The web configurator interface has been enhanced based on user feedback to expedite learning and simplify OMNI Agent setup. For example: enabling CUDA, configuring camera visibility zones, binding licenses for Standalone mode, setting face identification thresholds, and more.
  • Fixed incorrect naming of identification fields far and frr in identification events and activities.

2.4.1 (04 Dec 2023)

What's New

  • Selected optimal settings for face detection and identification, allowing for processing a large number of video cameras with a dense flow of faces.

Bug Fixes

  • Mask detection has been disabled due to quality instability and unwanted effects on performance.
  • Removed an error message that appeared in the console of the Standalone Agent prior to activation.

2.4.0 (16 Oct 2023)

What’s New

  • Added human activity recognition (HAR): falls, fights, lying and sitting. Now you can build business solutions to identify situations in which people need help.
  • Added OMNI Agent installation without GUI, which allows you to install it on server machines, as well as remotely via ssh;
  • Added people tracking on a floor map. This feature allows you to build high-level business intelligence (heat maps of people flows, people coming close to potentially hazardous areas, or getting into advertising areas, etc.).
  • Added the ability to recognize and track people who get into the regions of interest (ROI) or cross the lines marked on the frame from camera. Now you can organize perimeter security without fences and other physical barriers, control people coming close to the cliffs and bridge railings, react to people in certain camera areas for advertising purposes, etc.
  • Added the ability to deploy OMNI Agent in an isolated environment as a standalone application without connection to OMNI Platform. This feature allows you to use the API in third-party business solutions in the silhouette and face detection mode (without identification using OMNI Platform), reducing the cost of maintaining the server part.
  • Support for the ARM64 platform including NVIDIA Jetson (for face detection and tracking). Now it's possible to run the application on hardware with more computing power.
  • The installer has been reduced by half, as support for 11v1000 and earlier facial recognition methods has been discontinued. Faster downloading and saving of disk space after installation.
  • Added collection of "traces" (detailed logs) to analyze the results of video processing and identify problems.

Bug Fixes

  • Body detection pipeline now uses no more than 2 cores per thread instead of using all available ones. Now hardware resources are used more efficiently.
  • Fixed a rare overflow of the internal queue of OMNI Agent modules on slow machines. As a result, it was possible to optimize the use of hardware resources and improve the stability of OMNI Agent.
  • Fixed flickering body detections in the preview so that the user does not have a feeling of instability of the functionality.
  • Fixed rare OMNI Agent crashes.

2.3.3 (10 Oct 2023)

Bug Fixes

  • Fixed performance problems when using a large number of cameras.

2.3.2 (15 Sep 2023)

What’s New

  • OMNI Agent uptime is now shown in the web interface. Now you can find out about instability caused by external factors about which OMNI Agent cannot generate an error;
  • Guaranteed delivery of events when OMNI Agent works in networks with an unstable connection.

Bug Fixes

  • ISO 8601 timestamps in activities and events now use the same time zones (UTC +0);
  • Fixed virtual memory leak on MS Windows when OMNI Agent is used with large databases (from 100K);
  • Fixed false positive detections when people move in the field of view of several cameras.

2.3.1 (11 Aug 2023)

Bug Fixes

  • Fixed a bug where the returned score could be greater than 1.

2.3.0 (05 Jun 2023)

What’s New

  • Added generation and transmission of identification/non-identification events;
  • Added score setting for face recognition;
  • Agent settings are updated without restart.

2.2.0 (20 Apr 2023)

What’s New

  • Added multi camera support: now you can connect additional cameras to OMNI Agent;
  • Added support for FaceSDK recognition algorithms: method12v30, method12v50, method12v100, method12v1000;
  • Added CUDA 10 support.

Bug Fixes

  • Fixed rare hangs during shutdown;
  • Fixed flickering attention detections in OMNI Agent web interface.

Improvements

  • Added asynchronous database drop after OMNI Agent deactivation;
  • Added display of OMNI Agent version in web interface and via IP address of device with Agent:port/about;
  • Changed the appearance of the displayed face bbox.

2.1.1 (01 Feb 2023)

Bug Fixes

  • Fixed a bug with Agent intallation to non-system partition on Windows.

2.1.0 (23 Nov 2022)

Bug Fixes

  • Stable processes UUID;
  • Hangs during shutdown sequence when Agent is reconfigured the second time before its restart;
  • Rare database synchronization issues causing partial corruption of profile data.

Improvements

  • Added ability to transfer best shot via webhook;
  • Added ability to disable transfer of biometric template;
  • Better installer package compression;
  • Added ability to transfer best shots in original quality.

2.0.0 (10 Oct 2022)

What's New

  • FaceSDK 11v1000 support;
  • New database fetching algorithm.

Bug Fixes

  • Fixed several installation issues;
  • Signed installers of on-prem agent;
  • Fixed random changes (between startups) of web configurator's port when default is already in use;
  • Fixed (rare) random hangs on MS Windows during the shutdown sequence;
  • Fixed random shutdowns for cases of RTSP video streams with timed connection interruptions.

Improvements

  • Partial runtime reconfiguration support (internal usage only);
  • Response time optimizations;
  • Smooth synchronous preview;
  • Template version synchronization.