Intelligent Access Control System Based On User behavior youtube.com/watch?v=W3rJVaBky9Y CIVABIS Matjaž Gams Boštjan Kaluža, Erik Dovgan Jožef Stefan institute, Slovenia
Presentation Motivation Experimental environment Entry events Architecture Modules Integration Verification Discussion
Motivation (security project) Terrorist attacks – bypass sensors Malitious employee – drunk, angry... intercept unusual events based on intelligent experience 2 people entering, one registered employee “afraid”
Experimental environment Door sensor Card reader Fingerprint reader Camera
Entry event 1)Card identification 2)Fingerprint verification 3)Door opens 4)Door closes Unusual behavior ̴ 10 additional scenarios in advance Bomb attack – only door opens A terrorist steals a card and a finger
Architecture
Access sensors and Time&Space software Card reader Fingerprint reader Door sensor Time&Space controller Intelligent system Camera Camera module Videos TCP/IP ODBC
Module 1: Expert system A set of ̴ 10 predefined types of rules Verifies if the events are “legal” None of user behavior learning is used Examples of generic rules: 1)alarm / warning if event between time1 and time2 2)alarm / warning if more than N events in time 3)alarm / warning if no exit before time 4)alarm / warning if no exit in time
Module 2: Micro learning Learns user behavior on micro level – micro timing Algorithm: Local outlier factor Classification and explanation
Module 3: Macro learning Learns user behavior on macro level – macro timing / classification and explanation
Module 3: Vision Learns user behavior from video
Integration Regular eventAlarm event Main thread Expert systemMicro learningMacro learningCamera Displaying final result
Explanation
Measurements Our tests with our employees Our “simulated” tests with our employees Joint tests by security experts perform several of them
“Simulated” Measurements Tested modules: Expert rules, micro learning and macro learning Create regular accesses: Five people, each 40 learn and 10 test accesses – Create irregular accesses: Fake-identity experiment – generate entries with identification card of another person
Measurements - results okwarningalarm rules100%0% micro98%2%0% macro90%10%0% together88%12%0% okwarningalarm rules100%0% micro36%15%50% macro14%25%62% together13%18%69% Statistic for regular accesses Statistic for irregular accesses Ok – 88% of regular accesses Alarm – 69% of irregular accesses
Conclusion Designed and tested an original ambient- inteligence system for entry control based on user behavior It integrates arbitrary (currently four) independent modules and sensors Significant increase in security Patent pending, real-life application