Abstract : According to the world health organization, millions of elderly suffer from falls every year. These falls are one of the major causes of death worldwide. As a rapid medical intervention would considerably decrease the serious consequences of such falls, automatic fall detection systems for elderly has become a necessity. In this paper, an efficient machine learning-based fall detection algorithm is proposed. Thanks to the proposed local binary features, this algorithm shows a high accuracy exceeding 99% when tested on a large dataset. In addition, it enjoys an attractive property that the computational cost of decision-making is independent from the complexity of the trained machine. Thus, the proposed algorithm overcomes a critical challenge of designing accurate yet low-cost solutions for wearable fall detectors. The aforementioned property enables implementing autonomous, low-power consumption wearable fall detectors. © EURASIP 2018.
