Abstract
Automatic recognition of human activities is one of the basic techniques required for monitoring in security systems. In proposed method, following three kinds of human states in daily life can be recognized from measuring thermal images: resting state, active state, and absent state. These states are judged by comparing intensity change of human region among consecutive images. First, the active state was roughly determined as the region where the intensity changes. Then, each region was checked more precisely by Fuzzy reasoning based on degree change in intensity, average intensity of its area and its terminal area, and complexity of the boundary shape. Human region was determined by this process. On the other hand, the resting state is judged when the intensity doesn't change from previous human region. If the measured image belongs neither resting state nor active state, it was judged as absent state. The accuracy of the judgement of living states was 95% in a room arranged for an experiment. Judgement was miscarried in case that human area was small or overlapped with large and high temperature objects. Calculation time per one image was about 20 second, using a personal computer.
