A Comparison of Skin Detection Algorithms for Hand Gesture Recognition

Abstract

Hand gesture recognition software is becoming more accessible with the advances in depth cameras and sensors, but these sensors are still expensive and not freely available. A real time Hand Gesture Recognition software is designed to work with a low cost monocular web camera. Skin detection and skin extraction is a common form of image processing used for gesture recognition. A comparison of three different skin detection algorithms is performed. The three algorithms are: YCbCr thresholding, RGB-H-CrCb thresholding and KNN Classification. The results obtained for each algorithm show that the algorithms are unreliable with a low mean and a large standard deviation. It was concluded that the uncertainty of the accuracy of each algorithm reduces the effectiveness of the hand gesture recognition software and it is not implemented in the final design. Alternative skin detection algorithms are suggested to improve on the accuracies and latencies obtained.