| Abstrakt: |
Fall detection plays an essential part in the healthcare monitoring system and also helps the elderly and disabled people. In this paper, authors have proposed a methodology for an efficient detection of automatic fall, which can perceive every possible fall event by using human activity recognition (HAR). Eigen features and some other time related features are computed to the data, collected from the sensors associated with Android mobile devices. These features will be analysed to classify physical activities such as fall, walking, sitting, upstairs, downstairs, jogging, etc. In this work, long short-term memory (LSTM) Neural Network is used to classify human activities. Based on this, alerts will be generated in case of fall detection; otherwise, data will be archived for the future references. Performance of the proposed framework is evaluated on two benchmark datasets (WISDOM and UCI) and one real time tracked dataset. The accuracy of the proposed framework on tracked dataset is 91.48% and outperforms other classifiers. |
