This article aims to describe the fall detection problem corresponding to distress situations widely met with elderly persons and showing very heavy consequences on their physical and psychic health status. To solve the crucial issue of a correct abnormal event detection and the "doubt killing" issue facing potential waste of time and means for the emergency and rescue centers, Telecom SudParis developed data fusion approaches for heterogeneous and time-evolutive input data leading to specific processing architectures such as fuzzy logics and evidential networks based on the Dempster-Shaffer Theory of the belief. It also tends to show that these kinds of data fusion processing are encouraging for multi-sensors/agents healthcare decision process applied to the distress problem prevention.