Video feeds on Windows can now be configured and managed centrally for the entire cluster of SAFR Windows (and Linux) Platform machines. This means that a large deployment can be configured from a single machine using the Desktop Client (preferred) or the System Console. SAFR no longer requires video feed windows to remain open, nor do Windows users need to remain logged. SAFR will now also automatically resume processing on system reboot. This makes SAFR on Windows a fully resilient service that can handle power outages and be easily managed even when distributed on many machines.
To enable this, SAFR Windows Platform now comes with Virgo for Windows which performs video feed processing in the background. Windows Virgo supports Genetec, Milestone and Digifort VMS feeds as well as ONVIF, direct RTSP URL, and USB camera feeds. You can configure Windows Virgo via the Windows Desktop Client or the System Console. The Windows Desktop Client is recommended as a configuration tool in all cases and is required if configuring VMS feeds. When adding a feed simply select an auto-detected camera and choose operation mode.
As was already available on Linux, SAFR Windows Platform can now be configured for redundant DB operation. This means that all DB information (this includes face signatures, person meta-data and events but does not yet include images) will be stored in two or more separate machines and loss of one DB machine will automatically fail-over to another. Redundant DB operation also enables horizontal scalability of the face-matching operation which is distributed across all participating DB machines thus increasing size of deployment achievable (hardware estimator provides number of DB machines needed).
Keep in mind that you must have an odd number of DB machines for automatic failover to function and that the maximum number of redundant DB machines is 50.
SAFR can now be configured to synchronize watchlists from one SAFR Platform or Account to any number of other SAFR Platforms or Accounts. This means that SAFR Platform can now be deployed in a distributed manner with many independent SAFR Platforms at different locations and yet be kept updated with a watchlist maintained centrally (e.g. in Cloud).
You can configure SAFR Platform to synchronize one directory per account (tenant) from the System Console Status tab. Max latency for synchronization is 10 minutes and max throughput is ~20 records per second per sync connection. It might thus take up to 10 minutes to perform initial sync of 10K records.
DB matching speed and efficiency have been improved 5x. This means that matches are 5x faster and require 5x less processing power. This translates to significant TCO savings for deployments requiring large watchlists.
On single CPU core, 1 million faces can now be matched in 350-400ms.
SAFR Actions and Action Relay Event Service (ARES) now supports occlusion event attributes. This means you can configure actions to trigger specifically on occluded faces. For more information, search on "occlusion" in Action Relay Event Service - ARES manual.
Person detection balanced mode delivers 50% more throughput than max accuracy mode with only slight degradation in accuracy. This is now the default mode for person detection and is recommended for all cases when high accuracy of person body detection is needed (e.g. tracking in visually complex environments with several persons present).
In comparison, max speed person detection mode delivers 300% more throughput than balanced mode but with significant reduction in accuracy. However, this mode is commonly adequate for low complexity tracking such as casino tables or teleconferencing rooms.
SAFR Linux Platform now offers enhanced scalability across multiple NVIDIA GPUs. SAFR Linux VIRGO has been optimized to be even less reliant on CPU and to maximize use of NVIDIA GPUs. This means that a single large machine can support 6 NVIDIA T4 processors which amounts to a SAFR recognition payload of 90 1080p@15fps feeds or 75 4K@15fps feeds (inclusive of recognition).
This capability is also available in standalone VIRGO Ubuntu download from Developers page.
Person body detection and tracking is now enhanced with face recognition and thus takes on identity established through face recognition. As person body detection is more accurate than face (due to size and being detectable in nearly any orientation) this means that identity tracking with combined person body and face detection is more accurate than face alone. When more accurate account of identity presence before the camera is needed, person events can now be used which are augmented with associated face attributes.
This function is automatically enabled when both person (body) detection and face recognition are enabled.
This features previously introduced on Linux is now also available on macOS. It enables liveness detection based on consistent change in face orientation (pose) as an alternative to smile action. It can be used for walk-up and walk-through secure access scenarios that require liveness confirmation when paired with well positioned cameras.
SAFR native face detector is now multi-threaded on Android and offers higher frame-rate and accuracy than Google Vision face detector (available when Google Play is present on the device). The Android Mobile Client now delivers excellent face detection performance at ~15fps while utilizing 35% CPU and Google Pixel phone.
When video frame rate is higher than detection rate device can deliver, video frames will be appropriately skipped for analysis in order to not cause backlog of processing that would increase latency in detection and recognition.
Windows:
Android: