Deep Learning System for User Identification Using Sensors on Doorknobs.

Area of Science:

• Biometrics
• Human-Computer Interaction
• Security Systems

Background:

• Door access control is crucial for physical security.
• Current systems often rely on traditional authentication methods.
• Performance metrics like accuracy and speed are vital for access control systems.

Purpose of the Study:

• To investigate a novel approach for user identification in door access control.
• To utilize patterns of user interaction with a doorknob for authentication.
• To apply deep-learning algorithms to sensor data for behavioral biometrics.

Main Methods:

• Measuring user interactions with a doorknob using embedded accelerometer and gyroscope sensors.
• Applying deep-learning-based algorithms to analyze the collected sensor data.
• Evaluating identification accuracy across different user groups and sample durations.

Main Results:

• Achieved an overall user identification accuracy of 90.2% with 47 participants.
• User identification accuracy reached 97.0% for females and 89.8% for males.
• Demonstrated feasibility of identifying users with a short sample duration of 0.5 seconds (68.5% accuracy).

Conclusions:

• User identification through motor activity patterns is feasible for access control.
• This method offers a novel behavioral biometric for enhancing physical security.
• The approach provides an alternative to conventional authentication methods.