Explainable activity recognition for the elderly
F. Xavier Gaya-Morey, Cristina Manresa-Yee, José María Buades Rubio · 2023
In this work, we present the recognition of daily activities of elderly people at their home environment, as a way to help this collective preserve their autonomy. To do so, we have taken three different neural networks with near state-of-the-art performance for human activity recognition and trained them on a big dataset of elderly people activities at home environments: ETRI-Activity3D. Then, we have explained the results adapting LIME to work with video input, to further validate the models and to help future users understand the models’ predictions. In addition, we have separated the explanations in space and time as a way of reducing the required computation time. Our main objective is that through the deployment of our work on a real environment, abnormal (such as repetitions or omissions) or alarming (like falls) activities can be discovered and help elderly people accordingly.