Commentary Paper 2 on Action Signature: A Novel Holistic Representation for Action Recognition

This paper describes a method for action recognition using a classification algorithm based on mixtures of von Mises distributions processing action signatures. An action signature is a ID sequence of angles, forming a trajectory, which are extracted from a 2D map of adjusted orientations (subtracting the average orientation) of the gradient of the motion-history image. To obtain the action signature, the authors scan the image along the direction given by the average gradient orientation, selecting only the points for which the motion energy is equal to 1. The authors use Mixture of von Mises distributions to describe the action signature. The parameters of these distributions are found by applying the Expectation-Maximization (EM) algorithm. A similarity measure based on global alignment (inexact matching) and optimized by dynamic programming is used for the training and the classification of actions. To cope with the huge variability of actions, the authors adopt a learn-and-predict strategy in order to update and refine continuously the action clusters of the learned database. The algorithm is validated with a set of 25 videos. Each video portrays a person performing a sequence of 8 actions.