Wearable Sensor-Based Gait Analysis for Age and Gender Estimation

Area of Science:

• Biometrics and Human-Computer Interaction
• Wearable Sensor Technology
• Machine Learning for Healthcare

Background:

• Wearable sensors are increasingly used in healthcare for various applications.
• Automatic age and gender estimation from human gait is a significant area of research.
• Gait analysis using wearable sensors offers a unique biometric cue.

Purpose of the Study:

• To analyze and compare methods for age and gender estimation from sensor-based gait data.
• To evaluate the performance of different approaches in a competitive challenge setting.
• To identify the most effective techniques for gait-based human attribute estimation.

Main Methods:

• Utilized a large wearable sensor-based gait dataset comprising 745 subjects for training and 58 for testing.
• Collected gait data using three IMUZ sensors placed on the waist-belt or backpack.
• Analyzed 67 solutions submitted by ten teams, focusing on deep learning and conventional handcrafted methods.

Main Results:

• Deep learning-based solutions demonstrated superior performance compared to conventional handcrafted methods.
• The top-performing method achieved a 24.23% prediction error for gender estimation.
• The best age estimation achieved a mean absolute error of 5.39 years.

Conclusions:

• Deep learning approaches are highly effective for automatic age and gender estimation from gait.
• The study highlights the potential of wearable sensor data for biometric applications.
• Angle embedded gait dynamic images and temporal convolution networks show promise for accurate estimation.