Human Activity Recognition Under Partial Occlusion

Abstract

One of the major challenges in Human Activity Recognition (HAR) using cameras, is occlusion of one or more body parts. However, this problem is often underestimated in contemporary research works, wherein training and evaluation is based on datasets shot under laboratory conditions, i.e., without some kind of occlusion. In this work we propose an approach for HAR in the presence of partial occlusion, i.e., in case of up to two occluded body parts. We solve this problem using regression, performed by a deep neural network. That is, given an occluded sample, we attempt to reconstruct the missing information regarding the motion of the occluded part(s). We evaluate our approach using a publicly available human motion dataset. Our experimental results indicate a significant increase of performance, when compared to a baseline approach, wherein a network that has been trained using non-occluded samples is evaluated using occluded samples. To the best of our knowledge, this is the first research work that tackles the problem of HAR under occlusion as a regression problem. © 2022, Springer Nature Switzerland AG.