Fuzzy logic and differential evolution-based hybrid system for gesture recognition using Kinect sensor

The paper introduces a novel approach to gesture recognition aimed at physical disorder identification capable of handling variations in disorder expressions. The gestures are captured by Microsoft's Kinect sensor. The work is segmented into four main parts. The first stage describes a ‘relax’ posture through four centroids depicting four portions of the skeletal structure. In the second stage, when the subject is showing symptoms of any one of the 16 physical disorders, then the skeletal structure distorts; the bilateral structure is lost, and concept of ‘centroid’ computation does not seem relevant. Hence, in the second stage, ‘motion points’ depicting shifted centroids for the distorted posture are computed by distance maximization with respect to the four corresponding centroids obtained for the relax posture. This process is carried out by adapting the weights assigned to each joint by differential evolution. In the third stage, eight features are figured out on the basis of Euclidean distances and angles among the motion points of the distorted gesture. In the final stage, gestures are recognized using an interval type‐2 fuzzy set‐based classifier with 91.37% accuracy.