WIT Press

Reliable Situation Recognition Based On Noise Levels


Free (open access)








505 kb

Paper DOI



WIT Press


U.-P. K¨appeler, A. Gerhardt, C. Schieberle, M. Wiselka, K. H¨aussermann, O. Zweigle & P. Levi


Situation recognition based on audio signals can be used to determine situations in a meeting room while protecting the privacy by recording and analyzing only the noise level instead of the complete audio signal. A reliable situation recognition is normally obtained using Bayesian networks which do not only rely on context information itself but additionally on corresponding probabilities. Especially when the situation recognition itself should output a quality rating to the determined situation it is necessary that each analyzed information about all preconditions is rated with qualities or probabilities. This leads to the need of a conversion from uncertainty to probabilities when using sensor data to observe the environment to recognize situations. In this paper we compare different methods for situation recognition based on noise level measurements. We implemented a Multilayer Perceptron Neural Network, a simple empiric method and we describe the advantages of our method based on multinomial logistic regression which we adapted to a reliable and easily configurable situation recognition based on sensor data. To evaluate the methods we distinguished between the situations meeting, work and silence and recorded thousands of noise levels to calibrate and compare the different methods against each other. Our approach using logistic regression can be used as situation recognition based on sensor data. It is not necessary that the input information contains quality ratings but the system has to be calibrated with sensor data that can be assigned to all the situations that have to be distinguished. Since the result of this method contains probabilities to all situations it can also be used to analyze sensor data related to single preconditions of complex situation recognition algorithms based on Bayesian networks and different types of context information. Keywords: sensor, situation recognition, probability, logistic regression, quality, context.


sensor, situation recognition, probability, logistic regression, quality, context