An approach to the fusion of probabilities of activities for the robust identification of activities of daily living (ADL)

摘要

This paper describes how the precision of a system, which detects the activities of daily living (ADL), can be increased using the fusion of different sensors. Since many of the same activities can be detected by different sensors simultaneously, the use of fusion is predestined to increase the overall accuracy. The fusion can be used at various points in the data flow. If the fusion is used in the data flow at an early stage less information is lost, but higher effort is necessary for the implementation. In a first approach it is investigated how well the data at the highest level can be fused. At this level the classified activities can be found. The method Dempster Shafer was used to merge the uncertainty of data from different sources. The activity "knitting" could be recognized by the fusion with the data of a study significantly better than without the Fusion with the individual classifiers of the sensors (fusion: sensitivity = 83.5%, before: sensitivity = 32.5%).