Accuracy improvement of activity estimation methods using an inertial sensor-equipped cane

Abstract

We attempted to estimate the activity levels of elderly individuals using an inertial sensor- equipped cane. We estimated the activity levels of 20 elderly participants using machine learning techniques based on inertial sensor data and compared the results with those obtained from a commercially available activity tracker. As the results, it was found that the estimation accuracy was low in some cases. Then, we tried to introduce fine-tuning and classify the participants based on how they used the cane and estimation accuracy was improved. Therefore, we plan to further improve accuracy by considering the variations in cane usage.