Publication
Evaluating Deep Learning Models for Posture and Movement Recognition During the ABCDE Protocol in Nurse Education
Matthias Tschöpe; Stefan Gerd Fritsch; David Kariem Habusch; Vitor Fortes Rey; Agnes Grünerbl; Paul Lukowicz
In: 2024 International Conference on Activity and Behavior Computing (ABC). International Conference on Activity and Behavior Computing (ABC-2024), 6th International Conference on Activity and Behavior Computing, May 28-31, Nakatsu, Japan, Pages 1-10, IEEE, 2024.
Abstract
This work focuses on body posture recognition and motion detection within a self-recorded video dataset, specifically designed for nurse education concerning the ABCDE emergency protocol. We evaluate three categories of methods. The first category relies on 2D pose data extracted for each nurse from the videos. To classify the body postures and motions, we use either handcrafted features derived from the skeleton data or features that are automatically learned through convolutional neural networks (CNNs). The second category of models employs CNNs to recognize body postures and motions directly from the video data, without relying on skeleton data. In the third category, we apply our self-defined 1D-CNN to CLIP image embeddings that are extracted from the videos. The work concludes with a comprehensive evaluation and comparison of all the models presented.