A Gait-Based Real-Time Gender Classification System Using Whole Body Joints

Area of Science:

• Biometrics
• Computer Vision
• Machine Learning

Background:

• Gait-based gender classification is complex due to variations in walking style, speed, and joint occlusion.
• Existing research often focuses on specific joints, neglecting a holistic body approach.
• Utilizing the full body's joint data offers a more comprehensive gait analysis for gender determination.

Purpose of the Study:

• To develop and evaluate a machine learning model for gender classification using three-dimensional (3D) whole-body joint data captured by a Kinect sensor.
• To investigate the significance of all body joints in gait-based gender identification.
• To compare the proposed method's performance against existing image-based techniques.

Main Methods:

• Gait feature extraction from 3D joint positions recorded by a Kinect sensor.
• Statistical methods (Cronbach's alpha, correlation, t-test, ANOVA) for feature selection and validation of joint significance.
• Binary logistic regression model for gender classification using selected whole-body gait features.

Main Results:

• Cronbach's alpha indicated high joint reliability (99.74%).
• Statistical analyses confirmed significant differences between male and female joint movements during gait.
• All twenty joints were found to be statistically significant (p < 0.01) for gender classification.
• The proposed model achieved 98.0% accuracy in real-time gender classification using all body joints.

Conclusions:

• A novel machine learning approach effectively classifies gender using 3D whole-body gait data from a Kinect sensor.
• The study demonstrates the efficacy of analyzing all body joints for improved gait-based gender recognition.
• This 3D joint-based method surpasses the performance of traditional digital image-based gender classification systems.