**3. Materials and Methods**

An action recognition procedure is proposed. It is based on simple features extracted from the entire silhouette and its characteristic points, which are combined into action representation. The proposed approach applies our previous findings and recent research, aiming at improving the results presented in [29]. The dataset and the general processing steps are the same and will be explained in the following subsections. Several changes at the data preprocessing step are introduced, and a new parameter is added for the action representation method that previously yielded the highest accuracy.
