Automated directive fall detection system using single 3D accelerometer and learning classifier

Abstract

Technology advances to accelerate the quality and type of services provided for health care and monitor. Wearable sensor systems, composed of small and light sensing nodes, have the potential to revolutionize the health care system. An important application of wearable sensors can be the detection of fall with its direction, particularly for elderly or otherwise vulnerable people. In this thesis work, we implemented a direction-sensitive fall detection system prototype using a single three-dimensional accelerometer and machine learning algorithm which includes feature extraction and classification methods, e.g. PCA, SVM and KNN. Four types of fall, forward, backward, left and right falls are detected. In addition to the detection of a fall, it is also important to determine its direction, which could help locate joint weakness or post-fall fracture and help decrease reaction time. Most wearable fall detection algorithms are based on thresholds set by observational analysis for various fall types. However, such algorithms do not generalize well for unseen data sets and their applications in finding the directions of falls are not well recognized. A more appropriate approach, as presented in this thesis, is a machine learning based algorithm SVM and KNN were implemented for fall detection. Among the two methods, SVM provides better performances which leads to 96.45% of accuracy using PCA, mean and standard deviation features, exceeding the performances reported in the literature. The performances of the developed system in real time were also evaluated and they were found satisfactory. This work not only shows a machine learning algorithm that provides accuracy beyond the currently available algorithms but also shows direction-sensitive and cost-effective fall detection system using single 3D accelerometer.