MicroDoppler classification of human motions has been performed thus far without considering the Human ethogram. The ethogram is a catalog of possible activities and the way they are connected. For example, sitting and falling cannot be followed by walking without first standing. The same argument applies to previous motions, e.g., sitting can be only preceded by standing. From a motion classification perspective, the ethogram can be categorized into translation and in-place motions. Whereas the former mainly describe crawling and gait articulations, the latter are primarily associated with motions that do not exhibit considerable changes in range. In this paper, we exploit the human ethogram to guide and improve classification of activities of daily living. Using an FMCW radar with range and Doppler resolution capabilities, we compare the performance of the ethogram-based classifications with the case where all motion classes are considered all the time. The thrust of this comparison is not to advocate one type of human motion classifier over the other, but rather to show the impact of incorporating the ethogram sequence of human motion on classification performance.
|