In the frame of CompanionAble European project, we have utilized many ambient sensors in order to implement multimodal fusion application aiming to detect abnormal status of the care receiver CR. We have exploited actimetry and sound signal to track and localize the person in his living place. In order to increase localization accuracy and to improve performance of the tracking system, ultimodal fusion methods can be used to exploit several modalities. Among these techniques, we are nterested by those based on finite states model with a random variable that changes along the time. Relying on N-dimensional states transition matrix, we propose a fusion model of heterogonous data generated from multiple ambient sensors. To detect abnormal situations of the observed person, we would use a personalized matrix estimated by applying a Polya model.