Human gender recognition with upper body gait kinematics

Human gender recognition has captured the attention of researchers
particularly in computer vision and biometric arena. These investigations
based on computer vision or image processing have highlighted
applications in security systems, medical applications etc.
This work is primarily aimed at investigating a possible characteristic
difference in upper body movement kinematics between males
and females and, in the affirmative, if that kinematic information
alone is sufficient to distinguish each cohort from the other. We
use a Microsoft Kinect© to capture the human upper body gait
kinematics to uncover gender based kinematic variations.
Two groups of healthy volunteers (18 females, 16 males) were requested
to walk along a linear pathway in front of the camera.
Upper body movement kinematics were extracted from the male
and female cohorts during walking. Principal component analysis
(PCA) was employed to substantiate the differences between
the two cohorts in terms of kinematic information. Finally, we use
k-means clustering to classify and evaluate the performance of
the classification system. Despite of the limitation of the dataset,
e.g., the limited range of the Kinect© camera, the accuracy of the
proposed approach reached up to 94%, indicating that upper body
joint movements possess significant information content on human
gender based features.