Gait-Based Identification Using Deep Recurrent Neural Networks and Acceleration Patterns

Area of Science:

• Biometrics
• Machine Learning
• Human-Computer Interaction

Background:

• Biometric identification is crucial for security.
• Gait analysis offers a unique, non-intrusive identification method.
• Smartphone sensors provide accessible data for biometric applications.

Purpose of the Study:

• To develop a biometric identification system using smartphone accelerometer and gyroscope data.
• To optimize a recurrent neural network (RNN) for gait pattern recognition.
• To evaluate the efficiency and robustness of the proposed deep learning model.

Main Methods:

• Utilized smartphone accelerometer and gyroscope data from 15 users during gait activity.
• Pre-processed data to extract vertical acceleration patterns.
• Designed and implemented a deep recurrent neural network with LSTM cells for user recognition.

Main Results:

• Achieved user recognition precision exceeding 97% in most executions.
• Demonstrated the model's efficiency and robustness across different testing scenarios.
• Successfully learned individual gait features using the optimized RNN architecture.

Conclusions:

• The proposed deep neural network-based approach offers a highly accurate biometric identification system.
• Smartphone-based gait analysis using RNNs is a viable and effective method for personal identification.
• The system shows promise for real-world applications requiring secure and convenient user authentication.