
INCHEON AIRPORT
AI recognition-based next-generation airport immigration management system
Development time
7 Month
Manpower
4 Professionals
The Brief
AI vision recognition-based next generation airport immigration control system development and demonstration project promoted by the Ministry of Justice and Information and Communication Industry Promotion agency to implement deep learning-based CCTV data analysis system for detection and prediction of various crimes and terror attempts. Ellexi has been selected as a main AI technology developer for this project.
Service

Development AI module for detecting & tracing individual(s)
API based system integration
Maintenance


1

2

3

4

6
5
Data Collection & Modelling
Identification of specific person and development of tracking module using multiple cameras (3 or more)
Dashboard / Visualization


System should detect and track an individual(s) abnormal/dangerous behavior with multiple cameras in real time and should be able to integrate with existing security monitoring system.
Technological Challenges


Abnormal Behavior Recognition
System should detect and trigger alarm for any actions categorized as below.
Function
Challenges
2-person detection
Accuracy and processing time
-
Abnormal behavior detection in designated area)
Backward(reverse) movement
Accuracy and processing time
-
Abnormal behavior detection in designated area)
Make a dash
Accuracy and processing time
-
Abnormal behavior detection in designated area)
Leave belongings behind
Accuracy and processing time
Abnormal behavior, event, object tracking
-
Identification of abnormal behavior, accuracy, and processing time
-
Categorization of identified abnormal behavior.
-
Accuracy of object tracking



Road Map
-
Modelling module
-
Abnormal Behavior Explanation module





-
Detection module
-
Prediction
-
Module upgrades
Key Features
Abnormal behavior Detection
-
Multi-camera feature extraction
-
Specified/unspecified abnormal behavior detection
Reasoning function
-
Detection of more the five (5) simultaneous events
-
Specified/unspecified detection reasoning.
-
Facial recognition module integration capability
-
Operator Feedback function
Learning Features
-
Specific area learning
-
Specified/unspecified abnormal behavior detection learning features
Abnormal behavior description function
-
Multi-camera object extraction
-
Specified abnormal behavior classification visualization.
-
Object recognition-based abnormal behavior description display

The Result
Multiple-camera support: 2 or more
Multiple-camera object tracking: 3 consecutive cameras
Specified abnormal behavior detection accuracy: above 90%