Real-time hand gesture recognition for embedded system / Farah Farhana Mod Ma'asum, Suhana Sulaiman and Azilah Saparon

A system with low-cost hardware computer
webcam as the replacement of mouse click is being applied in this
research. In order to capture good image of hand in vision based
system, various segmentation techniques proposed by other
researchers are combined and tested to enhance the quality of
segmentation image. Canny edges and Otsu threshold technique
are used to segment the hand image while convex hull and
convexity defects algorithm are used to extract the image of hand
features. Embedded hardware (Arduino) board is employed for
validating the signal sent using hand gesture to replace LEFT
CLICK, RIGHT CLICK, MOVE cursors. An experiment is set
up to determine the accuracy in percentage of this work with ten
test subjects. They were prearranged for five minutes to become
familiar with the hand tracking system after the initial attempt.
The findings revealed that users are better trained in the second
trial after five minutes training. The results significantly
improved from 33.3 % to 52.6 % for LEFT CLICK, 46.7% to 61
% improvement for RIGHT CLICK while 56.7% to 77.3% for
MOVE cursor.