Wrist detection based on a minimum bounding box and geometric features

Wrist detection is a crucial element in the hand-pose estimation and hand-gesture recognition processes
in Human-Computer Interaction applications. Most methods use horizontal parallel lines to scan for the
location of a wrist line. The challenging problems in wrist detection are determining the orientation and
localising the horizontal parallel lines that scan for various hand poses. The proposed method automatically
detects a wrist, based on a minimum bounding box and geometric features. It also determines the
start and stop points to localise the scanning. The evaluation used a set of 1240 hand images with
ground-truth data taken from three sets of data. The hand images contained several gestures and individuals
to prove that the method is robust against various gestures. The evaluation shows that the method
successfully detects the image orientation and the wrist points with high accuracy.