Dynamic feature selection for online action recognition

Abstract

The ability to recognize human actions in real-time is fundamental in a wide range of applications from home entertainment to medical systems. Previous work on online action recognition has shown a tradeoff between accuracy and latency. In this paper we present a novel algorithm for online action recognition that combines the discriminative power of Random Forests for feature selection and a new dynamic variation of AdaBoost for online classification. The proposed method has been evaluated using datasets and performance metrics specifically designed for real time action recognition. Our results show that the presented algorithm is able to improve recognition rates at low latency. © 2013 Springer International Publishing.