A New Remote Health-Care System Based on Moving Robot Intended for the Elderly at Home

Bing Zhou, Kaige Wu, Pei Lv, Jing Wang, Gang Chen, Bo Ji, Siying LiuJournal of Healthcare Engineering 2018 · DOI

TL;DR

This paper introduces a novel remote health-care system for the elderly that leverages a mobile robot. It aims to bridge the communication gap for seniors, enabling them to connect with family and doctors, while also incorporating advanced features like automatic health data collection and fall detection. The system promises to enhance elderly care by providing a more accessible and comprehensive solution than current mobile-based approaches.

What Problem Does This Paper Address?

The paper tackles the growing challenge of providing effective remote health-care for an aging global population, particularly in China, which is experiencing a rapid increase in its elderly demographic. A significant issue is the difficulty many elderly individuals face in utilizing modern communication technologies like video calls due to hearing loss or limited mobility. This isolation can be dangerous, especially for those living alone who might experience sudden health emergencies. Existing remote care solutions often fall short, either being too complex for seniors, too expensive, or lacking crucial features like integrated health monitoring and intelligent environmental awareness.

Methods

The proposed system integrates several key functionalities onto a mobile robot platform (TurtleBot):

  • Voice Control: The elderly can control the robot's movement and initiate calls using voice commands.
  • Remote Control: Family members or caregivers can remotely control the robot and conduct video calls with the elderly via mobile phones.
  • Health Data Collection: The system incorporates sensors (e.g., heart rate sensor) to continuously collect health data, which is transmitted to the cloud for historical record-keeping and doctor access.
  • Target/Face Detection and Pose Estimation: This is a core innovation. The system uses a ResNet-based body-part detection model and Integer Linear Programming (ILP) to accurately detect and estimate the elderly person's posture. This enables:
  • Fall Detection: Automatic alarming when a tumble is detected.
  • Automatic Following: The robot can follow the elderly person.
  • Person Recognition: The system can recognize individuals, such as when an elderly person returns home.
  • Speech Recognition: The paper proposes a confidence model for speech recognition to improve accuracy and robustness, particularly in noisy environments or with non-standard speech patterns. They introduce a Genetic Algorithms Dynamic Time Warping (GA_DTW) approach for keyword determination, aiming for higher accuracy than traditional DTW.

Datasets

The paper mentions using the "human motion database of the Weizmann Academy of Sciences" for human posture estimation, specifically focusing on indoor scenarios. They also refer to experiments using a dataset of six human behaviors (walking, jogging, running, boxing, hand waving, and hand clapping) for simulation recognition. For object detection, they reference MAP (VOC2007) as a benchmark.

Key Insights and Ideas

  • Bridging the Digital Divide for Seniors: The core insight is that a physical robot, controlled by voice or remotely, can act as a more intuitive and accessible interface for remote health-care than relying solely on smartphones or computers.
  • Integrated Intelligence: The system's strength lies in its integration of multiple advanced technologies (voice, video, health monitoring, pose estimation, object detection) into a single, cohesive platform, addressing multiple facets of elderly care.
  • Proactive Safety through Pose Estimation: The ability to detect falls in real-time is a critical safety feature that goes beyond simple communication, offering a proactive approach to preventing serious injury.
  • Confidence in Speech Recognition: The emphasis on a confidence model for speech recognition is crucial for practical applications where perfect speech input is not guaranteed, especially with the elderly.

Key Results

  • Speech Recognition: GA_DTW achieved higher recognition rates than traditional DTW for both isolated words (95.07% vs. 89.09%) and continuous words (90.39% vs. 83.42%).
  • Pose Estimation: The paper presents comparison tables showing their method's accuracy against other algorithms for various human behaviors, indicating competitive or superior performance. Table 6 shows average position and joint point accuracy for different postures.
  • Target Detection: Table 7 shows that Faster R-CNN (used in their system) offers faster detection speeds and higher accuracy compared to R-CNN and Fast R-CNN.
  • Experimental Validation: The system was tested in real-world scenarios, demonstrating its ability to conduct video calls, remote control, and detect human behaviors. The authors claim excellent performance and good user experience.

Limitations and Future Work

  • Limited Vocabulary for Voice Control: While GA_DTW improves recognition, the system still relies on a defined set of keywords. The paper mentions the possibility of the speaker using words beyond the vocabulary.
  • Complexity of Pose Estimation: The paper acknowledges that pose estimation can be challenging due to self-occlusion and camera angles, especially with multiple people. While their ILP approach aims to mitigate this, it's an inherent complexity.
  • Hardware Constraints: The use of a TurtleBot implies certain limitations in terms of navigation and payload capacity, though not explicitly detailed as a limitation.
  • Future Work: The paper implicitly suggests further refinement of the speech recognition and pose estimation algorithms. The integration of more advanced health sensors and personalized care routines would also be logical extensions. The system's ability to learn and adapt to individual elderly users' needs and preferences is a promising area for future development.

Summary updated

Abstract

Nowadays, due to the growing need for remote care and the constantly increasing popularity of mobile devices, a large amount of mobile applications for remote care support has been developed. Although mobile phones are very suitable for young people, there are still many problems related to remote health care of the elderly. Due to hearing loss or limited movements, it is difficult for the elderly to contact their families or doctors via real-time video call. In this paper, we introduce a new remote health-care system based on moving robots intended for the elderly at home. Since the proposed system is an online system, the elderly can contact their families and doctors quickly anytime and anywhere. Besides call, our system involves the accurate indoor object detection algorithms and automatic health data collection, which are not included in existing remote care systems. Therefore, the proposed system solves some challenging problems related to the elderly care. The experiment has shown that the proposed care system achieves excellent performance and provides good user experience.
BibTeX
@article{zhou2018new,
  author = {Bing Zhou and Kaige Wu and Pei Lv and Jing Wang and Gang Chen and Bo Ji and Siying Liu},
  title = {{A New Remote Health-Care System Based on Moving Robot Intended for the Elderly at Home}},
  journal = {Journal of Healthcare Engineering},
  year = {2018},
  doi = {10.1155/2018/4949863},
}
Figure