In this paper, we discuss an event detection system using a wireless sensor network in the Ambient Assisted Living context. The sensors measure the environment in which the patients live, not vital parameters of the patient him- or herself, which is important in order to respect the privacy and informational self-determination of the patient. The major difficulties of the given setup with sensors in the environment are that the measurements are taken in an irregular fashion (as opposed to regular sampling) and that some of the sensors may be unreliable. To tackle these problems, we propose an event-detection framework that is based on the theory of conditional random fields [1]. We conduct experiments on real-life sensor data collected from a set of eight households. The experiments show that the conditional random field is well suited for ambiance surveillance.

System calibration of industrial computed tomography systems, used for dimensional measurement, is often based on scanning a reference object, measuring the deviations in the resulting tomogram, and using this information for later reconstructions [1, 2]. Because uncertainty is a factor in the process [3, 4], this calibration procedure itself is limited. In this paper, a simulation is used to determine the best case scenario in order to qualify lower limits of this procedure. The target is to analyse the measurability of the artefacts in the tomogram of the calibration object, independent from other influences.

Abstract. In this paper, we propose a method for the detection of irregularities in time series, based on linear prediction. We demonstrate how we can estimate the linear predictor by solving the Yule Walker equations, and how we can combine several predictors in a simple mixture model. In several tests, we compare our model to a Gaussian mixture and a hidden Markov model approach. We successfully apply our method to event detection in a video sequence.

No abstract