LifeTag

For lifesaving in shipwreck accidents, a wearable device, called LifeTag is designed for marine travellers. The LifeTag integrates localization, communication and life-sign detection modules, which will be triggered on automatically when falling into water and broadcasts the location and life status of the drowning people, so that rescuing ships within 10 nautical miles can receive the signal. This will speed up the drowning people searching and rescue process to improve the lifesaving probability. This paper focuses on the design of data processing technique to accurately detect the life status of drowning people. Real experiments are conducted which show that the inertial sensor data can be processed by machine learning method to efficiently detect the drowning people's life sign. But a challenge problem is that LifeTag requires a very efficient implementation of the classifier, which needs to be embedded into the resource limited firmware of the LifeTag device. To accomplish this, we investigate key feature selection and seek for the efficient and effective classifier design. A simplified online classifier is therefore investigated. Finally, we implement the optimized classifier into the firm ware. Practical experiments verify nearly 100% prediction accuracy of the proposed solutions.