TY - JOUR
T1 - Development of a methodology to predict and monitor emergency situations of the elderly based on object detection
AU - Youm, Sekyoung
AU - Kim, Changgyun
AU - Choi, Seunghyun
AU - Kang, Yong Shin
N1 - Publisher Copyright:
© 2018, Springer Science+Business Media, LLC, part of Springer Nature.
PY - 2019/3/1
Y1 - 2019/3/1
N2 - Because on the increase in the number of the elderly living alone and accidents occurring to them, the demand for a monitoring system capable of supporting fast response in case of an emergency situation by monitoring their everyday life in their residential spaces has been increasing. A framework and a system are presented to monitor the emergency situations of the elderly living alone using a low-cost device and open-source software. First, human pose recognition and emergency situations according to the pose change were defined using object recognition, and a procedure capable of detecting such situations was proposed. In addition, a pose recognition model was created using the TensorFlow Object Detection application programming interface (API) of Google to implement the procedure. Using a data preprocessing process and the created model, a system capable of detecting emergency situations and sounding an alarm was implemented. To verify the proposed system, the pose recognition success rate was examined, and an experiment on emergency situation recognition was performed while the angle and distance of the camera were varied in a setup similar to the residential environment. It is expected that the proposed framework for the emergency notification system for the elderly will be utilized for the analysis of various behavior patterns, such as the sudden abnormal behavior of the elderly, people with disabilities, and children.
AB - Because on the increase in the number of the elderly living alone and accidents occurring to them, the demand for a monitoring system capable of supporting fast response in case of an emergency situation by monitoring their everyday life in their residential spaces has been increasing. A framework and a system are presented to monitor the emergency situations of the elderly living alone using a low-cost device and open-source software. First, human pose recognition and emergency situations according to the pose change were defined using object recognition, and a procedure capable of detecting such situations was proposed. In addition, a pose recognition model was created using the TensorFlow Object Detection application programming interface (API) of Google to implement the procedure. Using a data preprocessing process and the created model, a system capable of detecting emergency situations and sounding an alarm was implemented. To verify the proposed system, the pose recognition success rate was examined, and an experiment on emergency situation recognition was performed while the angle and distance of the camera were varied in a setup similar to the residential environment. It is expected that the proposed framework for the emergency notification system for the elderly will be utilized for the analysis of various behavior patterns, such as the sudden abnormal behavior of the elderly, people with disabilities, and children.
KW - Emergency situation recognition
KW - Object-detection
KW - Pose recognition
KW - TensorFlow
KW - The elderly
UR - http://www.scopus.com/inward/record.url?scp=85053492998&partnerID=8YFLogxK
U2 - 10.1007/s11042-018-6660-7
DO - 10.1007/s11042-018-6660-7
M3 - Article
AN - SCOPUS:85053492998
SN - 1380-7501
VL - 78
SP - 5427
EP - 5444
JO - Multimedia Tools and Applications
JF - Multimedia Tools and Applications
IS - 5
ER -