Development of an agglomerative hierarchical clustering system for the identification of key patterns in hand regions

Abstract

Hand gesture recognition is an important topic in the field of Human Computer Interface. One of the key stages in a visual hand gesture recognition system is the detection of hand poses in the acquired video stream. This detection process can be very computational intensive due to the wide range of scales, orientations, and poses that a hand can present. All these image properties will be a disadvantage to the recognition of the different images due to the no-uniformity of the images. All the images presents its own properties of color, illumination, scale or position. To alleviate this problem, a fast detection of key sub-patterns of hand regions can be a potential solution. These sub-patterns should have a low dimensionality to allow a fast operation, and at the same time be enough distinguishable to identify the hand poses. For this purpose, this TFG proposes to develop an agglomerative hierarchical clustering system for the identification of key patterns in hand regions. Though different methods and metrics in the clustering process, it is possible to group the different hand image sub-regions in different clusters. Hand image sub-regions will be compactly represented by an appropriate feature descriptor. As a result, a cloud of feature vectors will be obtained in a same cluster. All hand image sub-region belonging to the same cluster obtained in the clustering process, will be identified to the representation of the different region of a hand.