Anomaly Detection for Civil Aviation Pilots Using Step-Sensors

To enhance the safety and efficiency of civil aviation, special attentions should be paid to pilot's physical and mental health. Existing works used video monitoring and social network mining to find the potential anomalies in pilot's daily life. However, video monitoring suffers from the privacy problems and social network mining is computational complex. To solve the problems of existing works, we propose a novel pilot anomaly detection method using step-sensors. The key idea of this method is that the pilots step information reflects their daily behaviors, and it is also influenced by the behaviors of the pilots social networks; if a pilot step number is extremely different from his historical step numbers or the step numbers of his social networks, this would probably be an anomaly. We, therefore, use the step-sensor to collect pilots step information and use the cluster method to detect anomalies. Experiments are held on 65 pilot candidates, which are divided into two social groups. We collect their step information during 50 days. Using our proposed anomaly detection method, outliers can be successfully detected for further analysis. Our method is also free of privacy problem and is highly efficient.