System

Main Processor

Two 16 C/32 T X86 processors, 2.1 GHz

GPU

Supports four NVIDIA TESLA T4 GPUs or six AIX3300-A GPU cards, or six AIC5000 Comparison Cards (the GPU card is optional, and the following functions are available with the GPU card and relevant software permission).

Operating System

CentOS Linux release 7.4.1708 (Core)

Memory

Eight 16 GB DDR4 memory modules with up to 24 slots

Disk

Five 3.5”4 TB HDD which can be expanded to maximum 32 TB (each HDD is 4T) with up to 8 slots.
7.2K RPM SATA 6 Gbps 512n 3.5”

Face Analytics

Face Detection

Detects and analyzes face image streams by gender, age, expression, glasses, mustache, face mask, opened and closed eyes, and opened and closed mouths.

Face Modeling

Supports extracting attributes from face images.

Face Arm

Compares the face snapshot against a designated face database to get information on the first person that exceeds the arming threshold.

History Alarm Records Query

Quickly search for history records on arming alarms to review information on alarms that were previously triggered.

Management of Registered Database

Manage and use multiple types of databases such as the blocklist database, and static database. You can add, delete and modify these databases and their members.

Search in Registered Database

Quickly search for registered database members by name, gender, date of birth and ID card number.

Search by Image in Registered Database

Images can be used to search for information. The results are compared with data on humans in the registered database, and the matches are displayed based on their similarity level.

Search in Snapshot Database

Conveniently search for history records of passing persons in the snapshot database.

Search by Image in Snapshot Database

Images can be used to search for information, and the search results can be filtered by time and channel. The results are compared with data on humans in the snapshot database, and the matches are displayed based on their similarity level.

1V1

Supports 1V1 face comparison, and returns similarity result.

Cluster

Supports clusters (when it is set to work with the video cloud platform)

Face Analytics Performance

Image Stream Analysis Capability

One T4 card supports up to 200 face images/s.
One AIX3300-A card supports up to 200 face images/s.
The actual performance depends on the software that the server works with.

Blocklist Database Arm

One T4 card supports processing a blocklist database with 5 million pieces of data, and can trigger real-time alarms for 100 face attributes/second.
One T4 card supports processing a blocklist database with 2 million pieces of data, and can trigger real-time alarms for 200 face attributes/second.
One T4 card supports processing a blocklist database with 1 million pieces of data, and can trigger real-time alarms for 200 face attributes/second.
One T4 card supports processing a blocklist database with 500,000 pieces of data, and can trigger real-time alarms for 300 face attributes/second.
The arming threshold is ≥80%
One AIC5000 card supports processing a blocklist database with 5 million pieces of data, and can trigger real-time alarms for 240 face attributes/second.
One AIC5000 card supports processing a blocklist database with 10 million pieces of data, and can trigger real-time alarms for 180 face attributes/second.
One AIC5000 card supports processing a blocklist database with 30 million pieces of data, and can trigger real-time alarms for 100 face attributes/second.
One AIC5000 card supports processing a blocklist database with 50 million pieces of data, and can trigger real-time alarms for 80 face attributes/second.
The arming threshold is ≥80%
The actual performance depends on the software that the server works with.

Search by Image

One server supports up to 100 million face data (30 million in registered database and 70 million in snapshot database by default) Search by image and respond in seconds (only in the case of working with video cloud platform, and exclusive use of hardware resources of single server)
One AIC5000 card supports up to 100 million face data (30 million in registered database and 70 million in snapshot database by default) Search by image and respond in seconds
The actual performance depends on the software that the server works with.

1V1

One T4 card supports up to 75 times/s
One AIX3300-A card supports up to 75 times/s
The actual performance depends on the software that the server works with.

Vehicle Analytics

Vehicle Recognition Mode

Recognizes the front and back of vehicles (supports recognition of license plates from Brazil and Middle East, and configuring license plate recognition for other countries).

License Plate Color Recognition

White, black, yellow, blue, and green

Vehicle Color Recognition

Recognizes a variety of colors such as white, orange, pink, black, red, yellow, gray, blue, green, silver, brown and purple

Vehicle Type Recognition

Supports large bus, heavy truck, medium truck, sedan, van, truck, medium bus, SUV, MPV and pickup

Vehicle Attributes Recognition

Recognizes vehicle attributes such as tissue boxes, sunshields, pendants, perfume, cards and permits.

Safe Driving Behavior Detection

For drivers, it detects when they are not wearing their seatbelt, calling while driving and smoking.
For co-drivers, it detects when they are not wearing their seatbelt and carrying a baby.

Target Detection

Supports vehicle recognition of incomplete snapshot

Image Stream Analysis Capability

One T4 card supports up to 3 million image/day (pixel from 2 MP to 9 MP)
OneAIX3300-A card supports up to 3 million image/day (pixel from 2 MP to 9 MP)
The actual performance depends on the software that the server works with.

Vehicle Brand/Logo Recognition

147 kinds

License Plate Recognition

License plate recognition with a minimum of 50 x 13 pixels is supported.
The actual performance depends on the software that the server works with.

License Plate Recognition Rate

With normal light in the day and normal fill light at night, resolution ≥120 × 120, clear, distinguishable license plate.
License plate accuracy: ≥96%.

Vehicle Type Recognition Rate

With normal light in the day and normal fill light at night, resolution ≥240 × 240, clear, distinguishable vehicles.
Vehicle type accuracy: ≥92%.

Vehicle Color Recognition Rate

With normal light in the day and normal fill light at night, resolution ≥240 × 240, clear, distinguishable vehicles.
Vehicle color accuracy: ≥95%.

Safety Belts Recognition Rate

≥85% (vehicle glass is clear)
The actual performance depends on the software that the server works with.

Recognition Rate of Driver and Passenger Phone Calling

≥80% (vehicle glass is clear)
The actual performance depends on the software that the server works with.

Search by Vehicle Picture

Images can be used to search for information, and the search results can be filtered by time and channel. The results are compared with data on motor vehicles in the database, and the matches are displayed based on their similarity level.

Search by Vehicle Picture Performance

Vehicle Search

One server supports up to 100 million data in snapshot database, searching by image and responding in seconds (only in the case of working with video cloud platform, and exclusive use of hardware resources of single server).
One AIC5000 card supports up to 100 million data in snapshot database, searching by image and responding in seconds
The actual performance depends on the software that the server works with.

Metadata Analytics

Target Classification

Classify targets including face, body, motor vehicles and non-motor vehicles. Generate intelligent structured analysis on the live video, history video (when using with the platform), and the video files that were uploaded offline, and then extract structured information on the moving targets. (for face, only image can be viewed without structured analysis)
.

Motor Vehicle Detection

Detects motor vehicles by a wide range of attributes such as type, color, brand and plate number (supports recognition of license plates from Brazil and Middle East, and configuring license plate recognition for other countries). Detects sunshield, not wearing seatbelt, calling, ornament (pendant and tissue box).

Non-motor Vehicle Detection

Detects pedestrians by a wide range of attributes such as gender, age, hair style, their top and bottom clothes and color, non-motor vehicle type and color wearing a hat, hat color, umbrella (canopy), umbrella (canopy) color, bag, bag color, shoe type and color, vest, mask color, raincoat, rearview mirror, truck, basket and direction.

Pedestrian Detection

Detects pedestrians by a wide range of attributes such as gender, age, hair style, their top and bottom clothes and color, wearing a hat, hat color, umbrella, umbrella color, bag, bag color, shoe type and color, vest, mask color, raincoat, cart, whether they are riding, and direction.

Real-time Display

Displays the analysis results in real time, and display humans, motor vehicles and non-motor vehicles in real time with tracking boxes.

Search by Attribute

Intelligent search of human, motor vehicles and non-motor vehicles by their attributes.

Search by Image

Images can be used to search for information, and the search results can be filtered by time and channel. The results are compared with data on humans, motor vehicles and non-motor vehicles in the database, and the matches are displayed based on their similarity level.

Video Recording Task

Generate intelligent structured analysis on history videos (when using with the platform), extract structured information on moving targets, and the video recording task can be accelerated automatically.

Local Task

Generate a structured analysis of your uploaded local video files, and extract information on moving targets, and the local task can be accelerated automatically.

Display Tracks on the Map

Yes (when using with the platform)

Computing Node Cluster

Supports clusters (when it is set to work with the video cloud platform)

Metadata Analytics Performance

Intelligent Video Stream

Supports real-time display of structured intelligent video streaming, including target attributes, and displays up to 5 items at the same time. Pedestrians (age, gender, top clothes and color, hat, umbrella and bag), vehicle (model, brand, license plate and color)

Metadata Analytics Performance

One T4 card supports up to 30-channel real-time 1080p video analysis of moving target or 100 images/s (1080p) metadata analysis of target or 3 million images/day (1080p) structured analysis.
One AIX3300-A card supports up to 80-channel 1080p video analysis of moving target or 100 images/s (1080p) metadata analysis of target or 3 million images/day (1080p) structured analysis.
The actual performance depends on the software that the server works with.

Search by Image Performance

One server supports up to 100 million data in snapshot database, searching by image and responding in seconds (only in the case of working with video cloud platform, and exclusive use of hardware resources of single server).
One AIC5000 card supports up to 100 million data in snapshot database, searching by image and responding in seconds
The actual performance depends on the software that the server works with.

Detection Rate of Moving Target

With normal light in the day and normal fill light at night, resolution ≥40×80, clear, distinguishable pedestrians and non-motor vehicles:
Detection rate of pedestrians: ≥95%;
Detection rate of non-motor vehicles: ≥95%;
With normal light in the day and normal fill light at night, resolution ≥120×120, clear, distinguishable motor vehicles:
Detection rate of motor vehicles: ≥95%.
The actual performance depends on the software that the server works with.

Pedestrian Recognition Accuracy

With normal light in the day and normal fill light at night, resolution ≥80×160, clear and distinguishable pedestrians:
Gender accuracy: ≥90%;
Hair style accuracy: ≥90%;
Clothes style and color accuracy: ≥90%;
Wearing accuracy: ≥90%;
Belongings accuracy: ≥90%.
The actual performance depends on the software that the server works with.

Recognition Accuracy of Non-motor Vehicle

With normal light in the day and normal fill light at night, resolution ≥80×160, clear and distinguishable non-motor vehicles:
Gender accuracy: ≥90%;
Hair style accuracy: ≥90%;
Clothes style and color accuracy: ≥90%;
Wearing accuracy: ≥90%;
Belongings accuracy: ≥90%;
Non-motor vehicle type accuracy: ≥90%.
The actual performance depends on the software that the server works with.

Vehicle Recognition Accuracy

With normal light in the day and normal fill light at night, resolution ≥240×240, clear, distinguishable motor vehicles:
Motor vehicle type accuracy: ≥92%;
Motor vehicle color accuracy: ≥90%;
Driver and passengers' safety belt accuracy: ≥85%;
Driver and passengers' phone call accuracy: ≥80%;
With normal light in the day and normal fill light at night, resolution width ≥120, clear, distinguishable number plate:
plate accuracy: ≥96%;
plate color accuracy: ≥90%.
The actual performance depends on the software that the server works with.

Port

Network Port

2 × 10000/1000 MB self-adaptive network ports

USB

2 × front USB3.0 ports and 3 × rear USB3.0 ports

VGA

2 × VGA ports

Others

1 × RJ-45 management network port

General

Power Supply

100–127/200–240 VAC, 50/60 Hz, 10/5 A

Power Redundancy

Dual

Power Consumption

≤ 800 W

Operating Temperature

+10 °C to +35 °C (+50 °F to +95 °F)

Operating Humidity

35%–90% (RH), non-condensing

Storage Temperature

–40 °C to +60 °C (–40 °F to +140 °F)

Storage Humidity

20%–93% (RH), non-condensing

Gross Weight

35 kg (77.16 lb)

Net Weight

27.5 kg (60.63 lb)

Product Dimensions

87.1 mm × 447.6 mm × 735.0 mm (3.43" × 17.62" × 28.94")(H × W × D)

Packaging Dimensions

273 mm × 754 mm × 1069 mm (10.75" × 29.69" × 42.09") (H × W × D)

Installation

Standard 19'' rack installation with guide rail

BTU

≤2729.7 Btu/h

Certifications

CE: SHEM190901702201ATC/SHEM1909017022AT
FCC: SHEM190901702301ATC/SHEM1909017023AT

Optional

Product Type

Hardware