A Hybrid Solution for Human Activity Recognition: Application to Wrist-Worn Accelerometry

This paper presents a systematic approach that is able to classify activities of daily living using a wrist-worn triaxial accelerometer. The strategy, which divides the multi-class problem into several binary classification sub-problems, is a mixture of both feature thresholding method and machine learning. The idea behind our design is to separate cyclic activities from transient ones, then to provide explicit recognition in these two categories separately. Experimental results have successfully proven the effectiveness of the proposed activity recognition process.