Biometric recognition through gait analysis

Area of Science:

• Robotics and Artificial Intelligence
• Biometric Security Systems
• Sensor Technology

Background:

• People recognition is crucial for collaborative robots and secure access control.
• Existing Red Green Blue Depth (RGBD) cameras have limitations including high computational cost, privacy concerns, and inability to recognize masked individuals.
• Gait analysis offers a potential alternative for biometric identification.

Purpose of the Study:

• To develop and evaluate BRITTANY, a biometric recognition tool utilizing gait analysis with LIDAR data and CNNs.
• To address the limitations of RGBD cameras in people recognition, especially in scenarios like the COVID-19 pandemic.
• To introduce and test a novel CNN architecture for classifying aggregated occupancy maps representing gait.

Main Methods:

• Development of BRITTANY, a system employing Laser Imaging Detection and Ranging (LIDAR) for gait data acquisition.
• Implementation of a novel Convolutional Neural Network (CNN) architecture for analyzing aggregated occupancy maps.
• Comparative analysis of the new CNN architecture against established models like LeNet-5 and AlexNet using identical datasets.

Main Results:

• BRITTANY demonstrated a successful Proof of Concept (PoC) in an indoor environment with five users.
• The novel CNN architecture effectively classified gait patterns from LIDAR data.
• The system achieved a final accuracy of 88% in people recognition.

Conclusions:

• BRITTANY provides an effective and accurate biometric recognition solution using LIDAR-based gait analysis.
• The developed CNN architecture shows promise for gait classification, outperforming traditional methods in this context.
• This approach offers a privacy-conscious and robust alternative for people identification in various applications.