Unsupervised Hand Detection in Class Room Using Combination of Skin Detection and Hough Transform

A large number of dataset can take a lot of effort in order to annotate the object that we observe, in this work is hand object. Therefore, we introduce an image processing method in order to perform unsupervised hand detection. The algorithm does not require any annotation data in order to do hand detection. Image processing step starts from skin detection in order to differentiate skin and non-skin region respectively. Subsequently, the border elimination was performed by specifying coordinates of each categories in dataset. Hand detection was performed by applying canny edge detector combined with hough transform in order to the hand coordinate. The proposed pipeline is validated by three categories of dataset. The result allows good accuracy rates of up to 97,677%.