A Novel Hybrid Fall Detection Technique Using Body Part Tracking and Acceleration

Falls by elderly individuals are a major issue in modern health care. A significant amount of research has been done in this domain. In this paper, we have proposed a hybrid solution for fall detection by using body part tracking and human body acceleration. The paper finds that in most cases vision-based fall detection systems work better and give a more accurate result when compared to non-vision-based systems because of the limitations of non-vision based systems (e.g., people forget to wear the wearable detection devices). The proposed system improves the accuracy of the state-of-the-art solution and reduces its computation cost. The vertical distances between head and body center, and human body acceleration are the features used in the proposed method and a Support Vector Machine (SVM) classifier is used to classify the outcome into two classes. The depth image from a Kinect Camera was used as an input to avoid any privacy issues that may occur by using RGB-based texture images, and the events were classified as an activity of daily living (ADL) or a fall.