Real-time hand gesture detection and recognition by random forest

Abstract

Detection and recognition of an unconstrained hand in a natural video sequence has gained wide applications in HCI (human computer interaction). This paper presents an unsupervised approach for the training of an efficient and robust hand gesture detector. Different with traditional hand feature descriptors, the proposed approach use pair-patch comparison features to describe the samples. And the random forest is introduced to establish a machine learning model. The pair-patch comparison features could rapidly describe a sample and the distributions of them have some similarity between the same classes. In the training procedure, a database which consists of a large number of hand images with corresponding labels and background images are established. Experimental results show that the proposed approach can achieve a detection and accuracy rate of 92.23% on the dataset. © 2012 Springer-Verlag Berlin Heidelberg.