[4K1-GS-9] Human interface:

[4K1-GS-9-01] Ensemble-based Robust Human Activity Recognition for Sensor Missing Scenarios

〇Yuto Nakamura¹, Fumiya Kitamori¹, Masamichi Shimosaka¹ (1. Tokyo Institute of Technology)

Keywords:HAR, ensemble, sensor missing

HAR(human activity recognition) using multiple sensor devices is being widely used in the daily application, and a system that can be used even if some of the devises are missing is required. Existing HAR models do not consider the sensor missing scenarios and the performance significantly degrades when some sensor devices are missing. Previous approaches for the sensor missing are to complement the raw data or features using active sensor data. However, these approaches need a large amount of computational resources in the devices and depend on the dataset-specific knowledge. Therefore we propose robust ensemble-based method, which does not depend on the datasets and tackle with the any combination of missing sensors. We conducted the experiment with two open datasets in HAR and verify the performance of proposed method overcomes the existing method in the sensor missing scenarios.