Real-time prediction of altered states in Drone pilots using physiological signals

This work describes a novel framework for the automatic monitoring of drone pilots' emotional state in order to recognize situations that can not be controlled at the moment of piloting a drone. The latter is achieved by real-time analysis of heart, breathing, temperature and sweating signals. The first step in our approach was to create a database of the signals mentioned before by following a customized acquisition protocol. Filtered methods are used to eliminate signal disturbances as well as feature extraction methods to reduce wave information and to generate the patterns that will provide a digital signature of the signals. Subsequently, neuro-fuzzy methods are trained with the database to fit a model representing a precondition of the pilot signals under normal circumstances, this is, with the pilot is not under stress or any other concern. Thus, for the recognition stage, the pilot signals are continuously sent to the trained model in order to produce a prediction of what the signal should look like, and this output is compared against the current pilot signals, the aim is to detect alterations in the pilot signal induced by situation of alert. These alterations are recognised and then translated into a stop command sent to the drone in order to avoid a possible collision due to erroneous controlling from the side of the pilot who is under stress or state of alert and may not react in time.