Viewpoint Insensitive Posture Representation for Action Recognition

Abstract

Human action recognition is a popular research area while it is changeling when facing various conditions related to viewpoint, subject, background, illumination and so on. Among all the variances, viewpoint variant is one of the most urgent problems to deal with. To this end, some view invariance approaches have been proposed, but they suffered from some weaknesses, such as lack of abundant information for recognition, dependency on robust meaningful feature detection or point correspondence. We propose a novel representation named “Envelop Shape”. We prove it from both theory and experiments that such representation is viewpoint insensitive. “Envelop Shape” is easy to acquire. It conveys abundant information enough for supporting action recognition directly. It also gets ride of the burdens such as feature detection and point correspondence, which are often difficult and error prone. In order to validate our proposed approach, we also present some experiments. With the help of “Envelop Shape”, our system achieves an impressive distinguishable result under different viewpoints.