Abstract
We propose a new intelligent fire judgment system using the time series of smoke density in early stage. We think that the time series is approximated by many small circles which is made by the set of three sequential points on the time series. In our system, 36-dimensional vector extracted the set of the circle is used as a feature vector, and a statistical modified Bayes discriminant function is used as a discriminant function. We evaluated experimentally efficiency of our method for actual 234 unknown time series. The correct judged rates are 100％ for flaming fire, 76％ for smoldering fire, 86％ for tobacco, 97％ for water mist, 78％ for cooking and the mean is 85.6％.
