Performance of Multi-directional MHI for Human Motion Recognition in the Presence of Outliers

This paper proposes a human motion recognition method which is robust against outliers. The basic motion history image (MHI) has been updated considering multi-dimensional history and energy images to calculate the feature vectors for recognizing various complex human motions. The ´rajio-taiso´ (Japanese Radio Physical Exercise) has been recognized by employing this method. This exercise was conducted by some nonprofessionals and hence the input video data has lots of disarray and variations along with the complex movements of the exercise. Moreover, some data contains outliers and noises. The recognition rate has been satisfactory in spite of the complex nature of the input data, outliers and noise, whereas, the basic motion history image [14] was performed with the video data by professional and formal environment.