In this thesis, we propose an activity recognition method using wearable sensors to prevent child home accidents. It is practically impossible to ask parents keep their eyes on babies 24 hours a day, 7 days a week. In order to prevent child home accident and reduce efforts of parents, new safety management method is required. Also early detection is needed to prevent unintentional injury at home. We analyze dangerous child activities, such as falls, poisoning, burns and electrocution, by utilizing statistical data sets and several observations. To recognize various dangerous activities, we attached only a single wearable device which includes a 3-axis accelerometer, an absolute pressure sensor, and a RFID reader to a child’s waist. The FFT analysis is adopted to extract features of the aggregated data, and SVMs are tested on these features. The overall accuracy of activity recognition using our proposed method was 95.42% with the SVMs.
