Dynamic textures for human movement recognition
Vili Kellokumpu, Guoying Zhao, Matti Pietikäinen · 2010
Human motion can be seen as a type of texture pattern. In this paper, we adopt the ideas of spatiotemporal analysis and the use of local features for motion description. We describe human movements with dynamic texture features. The proposed method is computationally simple and suitable for various applications such as action and gait recognition. We use Gentle AdaBoost to perform feature selection and build strong models from weak classifiers. We verify the performance of our methods on the challenging KTH and USF datasets, achieving high accuracy.