Intelligent Solution for Wireless Hybrid Facial Detection
and Recognition

Intelligent Solution for facial detection and recognition
ATIS facial recognition software
Technical Specifications
Face Detection
  • Detection of multiple faces in a photo
  • Detection of 70 facial feature points (eyes, eyebrows, mouth, nose, face contour)
  • Head rotation support: +/-30 degrees of in-plane rotation and +/-30 degrees out-of-plane rotation
  • Detection speed per CPU core:
    1. Real-time detection (webcam, +/-15 degrees of in-plane head rotation): 15 milliseconds
    2. Reliable detection (still images and videos, +/-30 degrees of in-plane head rotation): 80 milliseconds
  • Returned information for each detected face: position within image, age, gender
  • Introduced confirmation interval to deal with false acceptance rate and to improve overall accuracy
  • Age recognition accuracy depending on image quality and lighting conditions, the error rate is +/- 5 years
Face Matching
  • Enrollment time: 15 milliseconds
  • Biometric vector size: 1040 bytes
  • Matching speed per single thread: 150.000 / s
Multi-Core Support
  • IntellQ can be parameterized to use all available processor cores when executing face detection or recognition functions, to maximize the performance.
Real-life exercise
  • 10 CPU cores: 150 million face matches in 90 seconds
  • LFW (labeled faces in the wild) database 99.73% accuracy on 13.233 faces. This is maximum because the database has known
  • errors which are not removed otherwise it would be 100% accuracy
  • 15 million biometric vectors require 16 GB RAM for in-memory processing
  • Full HD real-time detection can be done by using only 1 CPU
HW Requirement
IntellQ’ System requirements
Minimum requirements

(Laptop or desktop)

  1. Intel Core i5 or i7; 4 cores
  2. 15.6″ LED Full HD display
  3. 16GB RAM
  4. 256 GB SSD
  5. Graphics full HD (e.g. AMD Radeon 520 2GB)
  6. Ethernet Gb LAN
  7. Wi-Fi 802.11ac
  8. Integrated HD camera
Speed & Scalability
  1. Maintains high performance matching against big databases.
  2. Self-selects optimal images.
  3. Maintains high accuracy despite bad light; hats; hoodies; beards;
  4. Glasses; scarves; partial face occlusion and more.
  5. Scalable from single user to the client-server collaboration system
  6. Multiple IP stream input with recording and face detection
Operating system:
  1. MS Windows 10 or higher