Human motion recognition using directional motion history images

Nowadays many persons are needed to observe images from surveillance cameras, because many surveillance cameras are installed in a town or in buildings. They are working under strain, because they must always watch the images from surveillance cameras to find a person with abnormal motion. Therefore reduction of the load is necessary. Many existing researches on human motion recognition are only recognition and the information of the result is not used. We assume that human motion is a set of basic motions. So, an overall motion can be understood using the recognition result of basic motions. The goal of the present research is to develop a method of human motion representation and translation using directional motion history images (DMHIs). In this paper, we describe a method of recognizing basic motions using the DMHIs. We perform the recognition by the Histograms of Oriented Gradients (HOG) feature. In the experiment, the number of bins and local area (cell) sizes for calculating the HOG feature are changed and the most suitable values are inspected.