What’s New in FindFace Multi 2.0

Enhanced Algorithms, UI, UX:

• New design: a brand-new design for fast and comfortable work is here.

  A completely reinvented UI provides smooth work with large amount of data: view object detailed information in sidebar, easily switch between opened tabs. Start exploring new interface with Web Interface Basics.

• Superior neural networks: Liveness technology that passed iBeta Level 2 testing.

  This FindFace Multi version presents a brand-new neural network model for mobile phones liveness.goodwin that passed iBeta Level 2 test. The model is aimed at silicon/latex masks and paper cylindrical face texture wraps detection. Also, FindFace Multi 2.0 includes a huge amount of brand-new neural network models or updated versions of previous neural network models with enhanced characteristics.

  See Configure FindFace Multi Neural Networks

Technical Changes:

• Technical architecture: microservices instead of monolith.

  See Architecture

• Docker-based distribution: FindFace Multi 2.0 runs as a set of docker containers described in a docker-compose file.

  See:

  • Docker Platform

  • Useful Docker Commands

• Vertical scaling support: more events can now be processed due to the better use of server capacity.

• Horizontal scaling support: process more events for less time by deploying new servers.

  Both vertical scaling and horizontal scaling support guarantee increased performance.

• Support of S3-compatible storage: an S3 storage provides reliable and long-term storage of an unlimited number of files and data. It gives a way to avoid the limitations of the file system when it comes to large amounts of data.

• CentOS support.

New Features:

• External VMS.

• PTZ camera control: the ability to rotate PTZ cameras via ONVIF is added to the video player control panel.

• VMS smart cleaner: configure video cleanup on a regular basis.

• Advanced webhooks.

• Active Directory integration.

• Episode and event feed and filtering in Video Wall.

• Episodes on the timeline in Video Player.

• Recognition of new object attributes.

  See:

  • Enable Face and Face Attribute Recognition

  • Enable Vehicle and Vehicle Attribute Recognition

  • Special Vehicle Recognition

  • Enable Body and Body Attribute Recognition

• New approach to attribute licensing: object attributes that were previously unlicensed or licensed as a group of several object attributes at once are licensed separately now.

Renamed:

• Interaction Analysis → Interaction Tracking; Cameras, Videos → Video Sources; Episodes, Events → Episodes & Events; Analytics → Audience analysis; Preferences → Settings; Cards → Record Index.

• car → vehicle

Fresh Neural Networks:

• Object detection

  • Face detection: a fresh face detection model face.jasmine_fast.003 with improved characteristics.

  • Vehicle detection: brand-new vehicle detection neural network models: car.gustav_accurate.004, used in findface-extraction-api and car.jasmine_fast.005, used in findface-video-worker.

  • Body detection: brand-new body detection neural network models: body.gustav_normal.015, used in findface-extraction-api and body.jasmine_fast.018, used in findface-video-worker.

  • 3-in-1 object detector: a new detector headbodyface.alpha000_normal.001 that returns three sub-objects belonging to a person: body/head/face.

• Object image normalization

  • Face image normalization: new face image normalization models facenorm.multicrop_full_center_size400 and facenorm.multicrop_full_crop2x_size400 that use a smarter normalization algorithm and a new version bee.v3 with improved characteristics.

  • Vehicle license plate image normalization: a new and much faster network model briacon.v0 and new versions anaferon.v5 and anaferon.v7 with improved characteristics for better vehicle license plate image normalization.

• Object recognition

  • Face recognition: fresh neural networks lime.v2 and mango_320 that ensure fast face recognition.

  • Body recognition: a brand-new model pedrec.clio with increased accuracy.

• Object attribute recognition

  • Head position recognition: a fresh neural network headpose.v2 to recognize head tilt/turn.

  • Liveness recognition: a brand-new model for mobile phones liveness.goodwin that passed iBeta Level 2 test. The model is aimed at silicon/latex masks and paper cylindrical face texture wraps detection. A brand-new neural network model liveness.web.v0 for webcam liveness spoofing attack detection. A new version liveness.pacs.v2 with improved characteristics.

  • Face quality recognition: a new version quality_fast.v1 that predicts face quality to get the best shot and filter trash.

  • Vehicle category recognition: a brand-new neural network carattr.categories.v0 that determines whether a vehicle belongs to any of these categories: motorcycle (including moped and scooter) or quad bike, car, truck, car with a trailer, truck with a trailer, bus, or articulated bus.

  • Vehicle weight and body size recognition: a brand-new neural network carattr.weight_types7.v0 that predicts a vehicle type according to its weight and body size.

  • Recognition of vehicle orientation: a brand-new neural network carattr.orientation.v0 to recognize vehicle view: front, rear, or side.

  • Special vehicle recognition: a new version carattr.special_types11.v1 that supports recognition of route transport, carsharing, gas service, military, and road service vehicles in addition to police cars, ambulance, rescue service, and taxi.

  • Vehicle license plate recognition: a new version carattr.license_plate.v7 that supports license plates of Czech Republic, Pakistan, Thailand in addition to Serbia, Lithuania, Latvia, Moldova, Estonia, Finland, Azerbaijan, Tajikistan, Turkmenistan, Mexico, Argentina, inherited from carattr.license_plate.v6 and carattr.license_plate.v5. Recognition of license plates of the UAE, Russia, Kazakhstan, Georgia, Saudi Arabia, Brazil, India, Uzbekistan, Vietnam, Belarus, Ukraine, Armenia, Kyrgyzstan, introduced in earlier versions of the product, is also supported, which gives a total of 27 supported countries.

  • License plate image quality: a new and much faster version carattr.license_plate_quality.v1, which is used to get the best shot of a vehicle’s license plate in a sequence of shots.

  • Silhouette-based age and gender recognition: a fresh neural network pedattr.age_gender.v0 to predict human’s age and gender by silhouette.

  • Bag presence recognition: a brand-new neural network pedattr.bags.v0 to recognize whether a person has a bag, a backpack, or a suitcase.

  • PPE recognition: a fresh neural network pedattr.protective.v1 to recognize the presence or absence of personal protective equipment and its color.

Functionality changes compared to 1.2:

• The Areas section is deprecated, but its functionality is still available through API. In the future the Areas functionality will be included into Counters. Consider using a previous version of FindFace Multi if you actively employ Areas functionality.

• Counter Chart for the last hour/day/week is removed from the counter settings interface: the chart functionality was not suitable enough for data analysis. Analysis of data, received from Counters, is available through report unloading or post-processing of data from API.

• Card Relations are no longer supported in the user interface, but the functionality still exists in API: being experimental, this feature failed to cover potential customers’ needs. It will probably be developed in the future.