IREX computer vision modules are optimized for different applications in Smart Cities. Trained against millions of images obtained in the CCTV networks, our neural networks deliver production-level accuracy in the real-world environment.
Weapon Detection People Falling or Laying Dangerous Crowd Density Intrusion Detection People on Railways Early Fire Detection Unattended Items Audio Analytics
Traffic Jams and Accidents Number Plate Recognition Vehicle Appearance (type, color, make and model) Driving Violations Average Speed Control Parking Management Tunnel Safety
Learn More
Open API
IREX connects third-party AI modules, IoT, access control, and other security sensors (e.g. metal detectors).
IREX provides pre- and post-incident intelligence to identify criminals and other subjects. IREX connects open data sources and leading intelligence providers to keep photo databases updated in your Secure Private Cloud:
Wanted Suspects
Criminals
Missing Persons
Sex Offenders
Terrorists
Sanctioned People
Algorithm Overview
IREX provides scalable software infrastrastructure to
Detect faces at any angle (en face, in profile, top/bottom views)
Build a robust descriptor from multiple images as the person moves
Recognize suspects in masks or with partial occlusion
Record the best shot of the person by analysing the angle of view, lighting, and motion blur
Monitor missing/wanted/staff lists
Search people by photo, person name, passport number, etc
Detect appearance features (gender, age, headdress, glasses, beards, mustaches, masks, skin color) for searching and traffic statistics
Detect spoofing for access control systems
NIST Face Recognition Vendor Test (FRVT) Ranking
IREX was ranked #1 in accuracy among all the U.S. companies in Face Recognition Vendor Tests (FRVT) for the Border and Kiosk Datasets conducted by the National Institute of Standards and Technology (NIST).
Border and Kiosk Photos are the most relevant datasets to identify missing people and suspects in non-cooperative mode from arbitrary city cameras. These datasets are less constrained in terms of viewing angle, lighting, and resolution when compared to the other FRVT datasets.
FRVT 1:N Identification Results [last updated: March 04, 2021]
Investigation Mode Rank One Miss Rate, FNIR(N, 0, 1) N=1.6M
Identification Mode HIGH T → FPIR = 0.001, FNIR(N, T, L) N=1.6M
Dataset
Border
Kiosk
Kiosk
irex_000 (US)
paravision_007 (US)
microsoft_5 (US)
trueface_000 (US)
0.0028
0.082
0.044
0.170
0.005
0.101
1.000
0.003
0.007
0.099
0.092
0.009
0.027
0.030
0.180
0.188
Dataset
irex_000 (US)
paravision_007 (US)
microsoft_5 (US)
trueface_000 (US)
Border
At left, rank 1 miss rates relevant to investigations; at right, with threshold set to target FPIR = 0.01 for higher volume, low prior, uses. Yellow indicates the most accurate algorithm. Smaller is better.
Count people against a tripwire in both directions
Count people for queue management
Estimate demographics (gender, age, skin color)
Export to Google Data Studio
Ethical Suspect Recognition
IREX has developed a set of tools and practices to ensure fair usage of its suspect recognition technology:
IREX is designed exclusively to search for pre-specified persons such as missing people, suspects, and opted-in staff members. It cannot be used to identify random people in public spaces
IREX has refined its permission management system to ensure that only authorized staff can access the suspect recognition data. This data is exclusively owned by the IREX client and is not shared with third parties.
IREX has implemented comprehensive logging to record user requests, including logins, photo uploads, search queries, people database, etc.
Privacy Protection
IREX customers have 100% ownership and control of their data
IREX uses commercially reasonable efforts to deploy and uphold the best security practices and standards
Traffic Management;Detect road congestion, accidents, hard braking, and shoulder stopping. Measure average speed, approximate speed (by means of video analysis), and exact speed (by means of radar integration).
Number Plate Recognition (NPR);Read and record plates of 30+ countries using preloaded templates (including single-line/double-line/standard/inverted plates). Read plates of other countries (by means of a universal reader). Search by vehicle number, owner name, and other data.
Vehicle Appearance; Classify passenger cars, vans, trucks, buses, bikes, and construction vehicles. Detect vehicle color.Detect vehicle make and model (under development). Search by vehicle appearance
Driving Violations; Detect stopping and parking violations (with a variable time threshold), double line crossing violations, and lane violations.Read and record number plates of violating cars
Parking Management; Detect vacant spaces