메인컨텐츠로 바로가기
Products

A·I based Algorithm

Face Recognition Algorithm

  • It is a highly trusted deep-learning based algorithm applied to various international airports and government complexes.
  • High processing rate thanks to excellent authentication performance and optimized library. It has an easily applicable API module that supports Windows / Linux / Embedded.

Features

  • High-speed face recognition based on deep-learning
  • Higher face recognition accuracy than the human eye.
  • Walkthrough face recognition available after passport scanning
  • More than 99% accuracy rate of identification
  • High success rate for 2-person detection just with face recognition
  • Minimalized deterioration of recognition accuracy under various environmental factors (e.g. low light intensity, backlight, etc.)
  • Compression of high-capacity image data
  • Flexible face-feature extraction
  • Solid performance regardless of lighting, posture, and facial accessories.
  • Libraries based on OpenCL(deep learning) or CUDA
  • High-capacity parallel signal processing algorithm based on GPGPU for effective feature points extraction and classification
  • Manycore architecture for resource management to minimize the power consumption in deep-learning operation
  • Multiple sensors converged to improve recognition performance and to reduce error Capable of extracting features from the video footages acquired by the CCTV cameras, through re-identification technique

References

More than 100 million passengers experienced the algorithm at Incheon Airport.
The same algorithm has been used in four government complexes of South Korea.

  • Government complex: Access control system based on face recognition in Seoul, Sejong, Gwacheon, and Daejeon complex
  • Incheon Int’l airport : eGates and their integration to the system
  • Chinggis Khaan Int’l airport in Mongolia and Vietnamese National Police : Immigration system
  • 행정안전부
  • 법무부
  • 인천국제공항공사
  • 법무부 출입국외국인정책본부
  • 경찰청
  • 국립중앙도서관

Specifications

  • Library API : C/C++
  • OS Environment : Windows, Linux, Embedded Linux
  • Module : DLL(Windows), so(Linux, Embedded Linux)
  • Extraction Rate : X86(0.45 Sec), Embedded(0.32 Sec)
  • Matching Rate : X86(500,000 pics/sec)
  • LFW Accuracy : EER Standard (more than 99.5%)