Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Human Emergency Detection during Autonomous Hospital Transports

Andreas Zachariae, Julia Widera, Frederik Plahl, Björn Hein, Christian Wurll

TL;DR

This work tackles safe autonomous hospital transports by equipping a mobile robot (PeTRA) with an RGB-D sensing setup and OpenPose-based pose estimation to detect emergencies during walking, rollator, and wheelchair transfers. The authors build a dataset of over 18,000 images and compare RF, MLP, and especially SVM classifiers, applying thresholding and a delay to maximize recall under imbalanced conditions, with SVM achieving 0.958 recall for walking and 0.622 for wheelchair. A AutoML baseline shows manual tuning outperforms automated approaches by about 5.3 percentage points in recall. The study provides a baseline for emergency detection in dynamic robot-assisted hospital transfers and shares a subset of the data for benchmarking, while highlighting opportunities for improvement in subtle wheelchair-emergency cues and end-to-end/video-based methods.

Abstract

Human transports in hospitals are labor-intensive and primarily performed in beds to save time. This transfer method does not promote the mobility or autonomy of the patient. To relieve the caregivers from this time-consuming task, a mobile robot is developed to autonomously transport humans around the hospital. It provides different transfer modes including walking and sitting in a wheelchair. The problem that this paper focuses on is to detect emergencies and ensure the well-being of the patient during the transport. For this purpose, the patient is tracked and monitored with a camera system. OpenPose is used for Human Pose Estimation and a trained classifier for emergency detection. We collected and published a dataset of 18,000 images in lab and hospital environments. It differs from related work because we have a moving robot with different transfer modes in a highly dynamic environment with multiple people in the scene using only RGB-D data. To improve the critical recall metric, we apply threshold moving and a time delay. We compare different models with an AutoML approach. This paper shows that emergencies while walking are best detected by a SVM with a recall of 95.8% on single frames. In the case of sitting transport, the best model achieves a recall of 62.2%. The contribution is to establish a baseline on this new dataset and to provide a proof of concept for the human emergency detection in this use case.

Human Emergency Detection during Autonomous Hospital Transports

TL;DR

This work tackles safe autonomous hospital transports by equipping a mobile robot (PeTRA) with an RGB-D sensing setup and OpenPose-based pose estimation to detect emergencies during walking, rollator, and wheelchair transfers. The authors build a dataset of over 18,000 images and compare RF, MLP, and especially SVM classifiers, applying thresholding and a delay to maximize recall under imbalanced conditions, with SVM achieving 0.958 recall for walking and 0.622 for wheelchair. A AutoML baseline shows manual tuning outperforms automated approaches by about 5.3 percentage points in recall. The study provides a baseline for emergency detection in dynamic robot-assisted hospital transfers and shares a subset of the data for benchmarking, while highlighting opportunities for improvement in subtle wheelchair-emergency cues and end-to-end/video-based methods.

Abstract

Human transports in hospitals are labor-intensive and primarily performed in beds to save time. This transfer method does not promote the mobility or autonomy of the patient. To relieve the caregivers from this time-consuming task, a mobile robot is developed to autonomously transport humans around the hospital. It provides different transfer modes including walking and sitting in a wheelchair. The problem that this paper focuses on is to detect emergencies and ensure the well-being of the patient during the transport. For this purpose, the patient is tracked and monitored with a camera system. OpenPose is used for Human Pose Estimation and a trained classifier for emergency detection. We collected and published a dataset of 18,000 images in lab and hospital environments. It differs from related work because we have a moving robot with different transfer modes in a highly dynamic environment with multiple people in the scene using only RGB-D data. To improve the critical recall metric, we apply threshold moving and a time delay. We compare different models with an AutoML approach. This paper shows that emergencies while walking are best detected by a SVM with a recall of 95.8% on single frames. In the case of sitting transport, the best model achieves a recall of 62.2%. The contribution is to establish a baseline on this new dataset and to provide a proof of concept for the human emergency detection in this use case.
Paper Structure (14 sections, 4 equations, 5 figures, 3 tables)

This paper contains 14 sections, 4 equations, 5 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Autonomous Hospital Transports
  3. Human Emergency Detection
  4. Dataset
  5. Implementation
  6. Method
  7. Training of recall-optimized classifier
  8. Decision threshold optimization
  9. Delay Time
  10. Results and Discussion
  11. Comparison with AutoML approach
  12. Evaluation results
  13. Limitations
  14. Conclusion

Figures (5)

  • Figure 1: Multi-mobility methods of PeTRA: Wheelchair transport (left), sensory coupling (center) and material transport (right).
  • Figure 2: Example images from the PeTRA Emergency Detection Dataset
  • Figure 3: Data processing pipeline from stereo camera to class output
  • Figure 4: Effect of threshold moving on ROC curve (left), Precision and Recall (right)
  • Figure 5: Effect of the delay on metrics (left), FPs and FNs (right)