Human Fall Detection Using Efficient Kernel and Eccentric Approach

Unintentional human falls are a very crucial problem in elderly people. If the fall goes unnoticed or undetected, it can lead to severe injuries and can even lead to death. Detecting falls as early as possible is very important to avoid severe physical injurious and mental trauma. The objective of this paper is to design the fall detection model using data of daily living activities only. In the proposed fall detection model, an eccentric approach with SVM based one-class classification is used. For the pre-processing step, fast fourier transformation has been applied to the data and seven features have been calculated using the preprocessed ADL dataset that has been calculated from the dataset of ADL (activities of daily living) activities acquired from the smartphones. An enhancement of the chi-square kernel-based support vector machine has been proposed here for classifying ADL activities from fall activities. Using the proposed algorithm, 98.81% sensitivity and 98.65% specificity have been achieved. This fall detection model achieved 100% accuracy on the FARSEEING dataset.