Human-skeleton based Fall-Detection Method using LSTM for Manufacturing Industries

According to the statistics of the Korea Occupational Safety & Health Agency, the incidences of falls in the manufacturing industry are increasing. In this paper, we introduce a fall-detection method based on skeleton data obtained from a 2D RGB CCTV Camera installed on the manufacturing floor. We proposed feature-extraction methods to improve of fall-detection accuracy and the construction of a fall-detection system using LSTM. Experiments were conducted through public datasets (URFD and SDUFall) to find feature-extraction methods that can achieve high classification accuracy. The experimental results showed that the proposed method is more effective in detecting falls than raw skeleton data which are not processed anything.