Intelligent Analysis AI Box Specification

  • Home
  • Intelligent Analysis AI Box Specification
AI Box Product View
Rizal March 4, 2024 0 Comments

Intelligent Analysis AI Box is an AI computing solution for edge computing requirements. The algorithms can be loaded and distributed in a flexible manner, and the design is founded on the idea of algorithm-defined hardware. You can construct a computing box that is tailored to your industry by loading several packages of industry algorithms. The ideal option for high-performance and cost-effective AI edge computing products is the integrated high-performance algorithm framework, which is made for edge-side AI deep neural network computing with strong arithmetic performance.

Features

1. Device Access

  • It can access high-definition network cameras that meet RTSP
  • It can Support H.265/H.264 video standard (Mac system does not support playback at present), up to 8 million pixel IPC access

2. Smart Applications

  • The total library supports up to 300,000 face pictures and supports 64 libraries
  • Support face dynamic bottom library: the recognition accuracy of face library will be improved, and
    the total capacity is 150,000 face;
  • Multi-type algorithm mixing: face mode, blend mode (face person binding + video structuring).
  • Video diagnosis: support a maximum of 4/8/16 video streams with full load
  • Face recognition: support a maximum of 4/8/16 video streams with full load
  • Video structured: Support a maximum of 4/8/16 video streams with full load
  • Perimeter alert: Supports a maximum of 4/8/16 video streams with full load
  • Behavioral vigilance: support a maximum of 4/8/12 video streams with full load
  • Item alert: Support a maximum of 4/8/16 video streams with full load
  • People counting: Support a maximum of 4/8/16 video streams with full load
  • Work Safety (industry extended algorithm package): support a maximum of 4/8/16 video streams with full load
  • Support face capture, face recognition comparison alarm, stranger recognition alarm, etc
  • Support the recognition of face attributes: gender, age, wearing hats, glasses, masks, etc
  • Support face, human body, motor vehicle, non-motor vehicle, face person binding, person-nonmotor vehicle binding-
  • Support the analysis of human attributes such as the color of people’s upper and lower clothing, the style of upper and lower clothing, the status of the backpack, and whether they wear a safety helmet
  • Support motor vehicle classification, color, brand, driving direction and non-motor vehicle
    classification and other vehicle attribute analysis
  • Support perimeter vigilance: vehicle stopping, vehicle leaving, personnel wandering, wall detection, intrusion, boundary crossing, battery car entering the ladder, climbing detection
  • Support behavior detection: running, falling, smoking, making phone calls, looking at mobile phones,
    sleeping on duty, leaving work, overstaffing on duty, less staff on duty, gathering of personnel, and scuffling of personnel , armed detection.
  • Support goods alert : sundry stacking, item guarding, and items left
  • Support people counting: regional people counting, entrance and exit people statistics
  • Support production safety supervision (industry extended algorithm package): safety helmets, uniforms, safety belts, reflective clothing, flame, smoke, fire protection facility detection, liquid leakage, Mask testing

3. Specification Parameters

System parameters

Main processor: High-performance embedded microprocessors
Operating system: Embedded Linux

Device access

Video stream input – Video Resolution: 1920 x 1080 (2 MP), 2560 x 1440 (4 MP), 3840 x 2160 (8MP)
Video decoding type: H.264/H.265

Working mode: multialgorithm parallel, configurable by channel

  • Video diagnosis (full load maximum 4/8/16 channels, 1 function per channel): screen
    abnormal detection
  • Face person binding + face recognition: (full load maximum 4/8/16 channels): face capture,
    face recognition, face attributes, human body capture, human attributes, face person binding.
  • Video structured: (full load max. 4/8/16 channels):
    – Image capture: face, human body, motor vehicle, non-motor vehicle, license plate
    – Attribute output: face, human body, motor vehicle, non-motor vehicle, license
      plate, license plate recognition
    – Association relationship: face person binding, car-license plate binding, personnon-      motor -vehicle binding-
  • Video inspection: (full load maximum 4/8/16 channels, 1 function per channel): screen
    abnormal detection.
  • Perimeter alert (fully loaded 4/8/16 roads, 4 functions per channel): vehicle stopping,
    vehicle leaving, personnel wandering, wall detection, intrusion, crossing the boundary,
    battery car entering the ladder, climbing detection.
  • Behavioral Alert (fully loaded 4/8/12 channels, 4 functions per channel): running, falling, smoking, making phone calls, looking at mobile phones, sleeping on duty, leaving work, overstaffing on duty, young personnel on duty, personnel gathering, personnel scuffling, armed and holding swords.
  • Goods alert (fully loaded 4/8/16 channels, 2 functions per channel): debris stacking, item guarding, items left.
  • People counting (fully loaded 4/8/16 roads, 2 functions per channel): regional people
    count, entrance and exit people statistics

Optional algorithms

Production safety supervision (fully loaded 4/8/16 channels, 2 functions per channel):
safety helmets, uniforms, safety belts, reflective clothing, flames, smoke, fire facilities
detection, liquid leakage, mask detection.

Alarms are reported

Support face capture, face recognition, face attributes, face man-machine nonlicense capture and attributes, license plate recognition results, and alert alarm
analysis results reporting.

Algorithm recognition accuracy

  • Face: face capture rate ≥ 99%, false capture rate < 1%, recognition pass rate: <
    99.5%, recognition false recognition rate: < 0.5%.
  • Human body: human capture rate ≥ 95%, false catch rate < 1%.
  • Motor vehicles: motor vehicle capture rate ≥ 90%, false catch rate < 1%.
  • Non-motorized vehicles: non-motorized vehicle capture rate ≥ 95%, false catch
    rate < 1%

Interface Parameters

  • Network interface: 2, 100M/1000M adaptive Ethernet, RJ45 interface
  • Audio output: 1
  • Audio input: 1
  • Front USB port: 1 x USB2.0 and 1 x USB 3.0 on the front Rear USB interface 2 x USB 2.0
  • Reset button:1
  • Power indicator (PWR): 1
  • Run indicator (RUN): 1

System Capabilities

  • Face recognition: Employee passage at the entrance; List control and identification of key personnel, alarm, stranger identification, etc
  • Video structuring: Capture, attribute analysis, license plate recognition of faces, human bodies, motor vehicles, non-motor vehicles, license plates, etc
  • Algorithmic warehouse management: Support uploading algorithm packages through web, API, MegConnect protocol, support algorithm package installation, uninstallation, deletion
  • Network protocols: TCP/UDP/HTTP/MULTICAST/DHCP/FTP/NTP/HTTPS/RTSP/GAT140, etc
  • Dual network ports: Supports three modes: multiple access setting, load balancing, and primary/standby mode
  • Log queries: It can query, search and display the capture information of face man and
    machine non-brand

Environmental Requirements

  • Operating temperature: -30℃ ~ +70℃
  • Storage temperature: -30℃ ~ +70℃
  • Relative humidity: 10% ~ 90%RH non-condensing

Other

  • Power supply: DC12V±10%,2A
  • Structure: Metal chassis
  • Dimensions: (L/D) 229×193×49(mm)
  • Weight: <2KG
en_US