Experimental studies and model development for individual-specific fall detection

Falls are a major cause of injuries and fatalities in daily activities and occupational settings. Fall detection has been suggested to be an effective fall prevention strategy. It can help initiate timely medical assistance for fall victims, and/or activate on-demand fall prevention systems (e.g. inflatable hip protectors) to prevent the physical injuries caused by fall impacts. The objective of this research was to develop a novel fall detection model based on the statistical process control chart. Given that a substantial proportion of falls result from slips, this research focused on detecting slip-induced falls. In order to achieve the research objective, three studies were conducted. The first study determined the appropriate fall indicators for fall detection research by experimentally examining a comprehensive set of kinematic measures. The body kinematic measures were compared between slip-induced falls and non-fall activities (i.e. normal walking and successful recovery after slips). Five kinematic measures were identified as the appropriate fall indicators since they were able to effectively and efficiently differentiate falls from non-fall activities, especially at the early stage of loss-of-balance due to slips. In the second study, a novel pre-impact fall detection model based on the statistical process control chart was developed and evaluated. The fall indicators were selected based on the experimental findings from the first study. The proposed fall detection model was individual-specific, since it was constructed using the individual historical movement data. The fall detection model demonstrated a high accuracy with sensitivity and specificity up to 94.7% and 99.2%, respectively. In addition, this model can also provide sufficient time for triggering fall protection device in the pre-impact phase, thus being potentially effective in preventing fall injuries. In the third study, a novel fall indicator defined by a linear combination of multiple kinematic measures was used in the proposed fall detection model. To specify the fall indicator, an optimization procedure was performed in which the trial and error method was used to determine the relative weightings of selected kinematic measures that were associated with the optimal fall detection performance. The sensitivity, specificity, and sum of squared errors of the fall detection model were 97.3%, 99.2% and 0.00133, respectively. The results were superior to those obtained when using fall indicators defined by a single kinematic measure.DOCTOR OF PHILOSOPHY (MAE