Resolving ambiguities in accelerometer data due to location of sensor on wrist in application to detection of smoking gesture

摘要

Diseases resulting from prolonged smoking are the most common preventable causes of death in the world today. Automated identification of smoking gestures can help to initiate the appropriate intervention method and prevent relapses in smoking. In previous work, we investigated the success of utilizing accelerometer sensors in smart watches to identify smoking gestures. Our experiments have indicated that identification of smoking gestures is indeed possible with 85%-95% accuracy through the use of Artificial Neural Networks (ANNs). As a follow-up study we present an investigation into the ambiguities in accelerometer data that arise due to the position of the smart watch on a person's wrist. As such, we have developed a method for transforming data to resolve the ambiguities for eight common configurations on the wrist. In this study we have utilized sensor data from the Pebble Time Steel smart watch. Results of our investigation indicate 100% success in recovery of individual smoking gestures after our developed transformation. Additionally our results indicate an average increase of 29.6% in detection accuracy when the method is applied to continuous smoking sessions. The methodology created in this work will be integrated into a mobile application for the automated detection of smoking to make it more flexible and robust. Inclusion of this method will greatly increase the accuracy of the ANN when it is faced with varying configurations of the watch.