Spatiotemporal recursive hyperspheric classification with an application to dynamic gesture recognition

摘要

The Recursive Hyperspheric Classification (RHC) is a novel algorithm well-suited for classifying noisy, multivariate datasets. Creating a taxonomy of hyperspheres, the RHC algorithm partitions a search space into labeled regions that are consumed in the recognition process of unlabeled data. While the algorithm is robust, classical RHC cannot cope with spatiotemporal data, such as dynamic gestures, because there is no mechanism that will facilitate temporal learning. Nonetheless, there exists a strong demand for computers to actively recognize gestures, for the simplest gesture can encode and convey an abundance of information. Gestures are a natural mode of communication by means of body movement, and they can imply intent. Therefore, in this paper, the Spatiotemporal Recursive Hyperspheric Classification (STRHC) algorithm is introduced, which utilizes a temporal queue, allowing the algorithm to classify and recognize temporal data, including human gestures that are sensed with the Xbox Kinect, a popular motion sensor. When validating its strength, STRHC has achieved up to a 95.33% average recognition rate while harnessing a diverse dataset of gestures.