Your Body Defines Your Fall Detection System: A Somatotype-Based Feature Selection Method
Xiaoling Fu, Zheng Tan, Chungang Yan, Zhong Li, Cheng Wang, Cheng‐Xiang Wang · 2017
Fall detection for the elderly is in great demand in order to mitigate the effect of falls. As fall detection system is a safety shield on which people's lives depend, high detection sensitivity is always the pursuit of fall detection system. Previous works prove that the sensitivity is correlated to the type of detection features. But generating personalized optimal detection features is difficult due to its high computation complexity. In this paper, we propose a somatotype-based feature selection method which can give user's optimal features without extra cost. Based on the finding that user's optimal detection features can be determined by their somatotype features (i.e., height and body mass index), we partition all users into different clusters according to their somatotype features and calculate the optimal features for each cluster. Several experiments prove that feature selection carried on somatotype based group can increase the detection accuracy effectively.