Abstract
Elderly sometime fall down from a bed and fractured their femur. Fall accident must be avoided to improve the quality of life. To solve this problem, we proposed monitoring system preventing fall down using DBN (Deep Belief Network) and Kinect.
However, there is a problem that the proposed system was not able to adapt individual difference of behavior leading to fall down from a bed for social deployment.
In this paper, it is proposed that monitoring system preventing fall down from a bed is adapted for individual difference of behaviors. Therefore, we proposed a learning method to adapt for individual difference of behaviors. In other words, distinctive behaviors are learned by monitoring system. The point of the discussion of this paper is behavior recognition of target. It does not the discussion about how to prevent fall down after dangerous behavior detected.
In the experiment, capability of the proposed learning method is evaluated.
You have full access to this open access chapter, Download conference paper PDF
Similar content being viewed by others
Keywords
1 Introduction
Elderly sometime fall down from a bed and fractured their femur. Fall accident must be avoided to improve the quality of life. According to the health care workers, the subject individual specifically behaves before monitored person fall out of bed. To solve this problem, we have proposed a system using Web camera [1,2,3,4,5]. In the dark room at night, the detection ability of previous system is low, because the brightness adjustment processing of a Web camera is not able to adjust brightness of the dark room. It is a problem that the previous system is not able to use in the dark room at night.
Therefore, we proposed monitoring system preventing fall down using DBN (Deep Belief Network) [6] and Kinect [7]. The purpose of this research is appreciating conceptually human behaviors. From the previous studies, it had obtained a high detection rate with respect to individuals. However, there is a problem that the proposed system was not able to adapt individual difference of behavior leading to fall down from a bed for social deployment.
In this paper, it is proposed that monitoring system preventing fall down from a bed is adapted for individual difference of behaviors. Therefore, we proposed a learning method to adapt for individual difference of behaviors. In other words, distinctive behaviors are learned by monitoring system. The point of the discussion of this paper is behavior recognition of target. It does not the discussion about how to prevent fall down after dangerous behavior detected.
In the experiment, capability of the proposed learning method is evaluated.
2 Behaviors of Monitored Person
In this research, the behavior of monitored person is classified into two states. One is safe behavior. Another is dangerous behavior. A state of lie on a bed is defined as a safe behavior. A state of all fours, sitting, flapping feet and flapping arm is defined as a dangerous behavior. Figure 1 shows examples of safe behaviors. Figure 2 shows examples of dangerous behavior. Figures 1 and 2 is measured by Kinect.
3 Monitoring System
3.1 Detection Processing
Figure 3 shows proposed monitoring system flow.
The proposed system is consisted with a PC (personal computer) and Kinect. Kinect, which measured distance between Kinect and monitored person, is developed by Microsoft. Target’s behavior is measured by Kinect and send a PC via USB (universal serial bus). The program of proposed system is running on a PC. The program converts measured data to input data of DBN and detects dangerous behaviors. A target’s body is extracted by the preprocessing using a threshold, which is set height of bed. The extracted data by preprocessing are normalized at one between two thresholds. The normalized data are inputted in DBN to recognize monitored person’s behaviors.
3.2 Proposed Learning Method
The proposed learning method is described as follows. First, initial learning is executed, in order to learn the variation of physique and basic behavior of targets. Second, user’s distinctive behavior is collected from the individual monitoring system. Where, the collected data is consisted of data suggested by user and monitoring system using learned DBN with initial leaning data. And, learning data is constructed for each user using a collected data. Finally, continuous learning is executed using constructed learning data for each user.
4 Experiment
The proposed learning method is evaluated. Five subjects participate in the experiment. Each five subjects behave distinctively on a bed. And, behavers of subjects are measuring by Kinect. DBN, that Initial learning is completed, is used for the continuous learning. Where, collected data by the proposed procedure is used for the continuous learning. From the experimental results are shown as follows. When the initial learning is completed, correctly rate of the dangerous behavior is 83.2%(208/250) and rate of the safe behavior is 84.8%(212/250) (Table 1). After proposed learning method is executed, correctly rate of the dangerous behavior is 81.6%(204/250) and rate of the safe behavior is 91.2%(212/250) (Table 2).
5 Conclusion
In order to prevent a fall down from a bed, we have been developing the monitoring system. However, there is a problem that the proposed system was not able to adapt individual difference of behavior leading to fall down from a bed for social deployment. To solve this problem, we proposed learning method of the monitoring system.
From the experimental results, ability of the proposed learning method is confirmed. It was confirmed that the monitoring system can adapt to the characteristic behavior of each individual.
References
Ikeda, R., Satoh, H., Takeda, F.: Development of awaking behavior detection system nursing inside the house. In: International Conference on Intelligent Technology 2006, pp. 65–70 (2006)
Matubara, T., Satoh, H., Takeda, F.: Proposal of an awaking detection system adopting neural network in hospital use. In: World Automation Congress 2008 (2008)
Satoh, H., Takeda, F., Shiraishi, Y., Ikeda, R.: Development of a awaking behavior detection system using a neural network. IEEJ Trans. EIS 128(11), 1649–1656 (2008)
Yamanaka, N., Satoh, H., Shiraishi, Y., Matsubara, T., Takeda, F.: Proposal of the awakening detection system using neural network and it’s verification. The 52nd The Institute of Systems, Control and Information Engineers (2008)
Satoh, H., Takeda, F.: Verification of the effectiveness of the online tuning system for unknown person in the awaking behavior detection system. In: Omatu, S., Rocha, M.P., Bravo, J., Fernández, F., Corchado, E., Bustillo, A., Corchado, J.M. (eds.) IWANN 2009. LNCS, vol. 5518, pp. 272–279. Springer, Heidelberg (2009). doi:10.1007/978-3-642-02481-8_39
Yoshua, B., Pascal, L., Dan, P., Hugo, L.: Greedy layer-wise training of deep networks. Adv. Neural Inf. Process. Syst. 19, 153–160 (2006)
Satoh, H., Shibata, K., Masaki, T.: Development of an awaking behavior detection system with kinect. In: Stephanidis, C. (ed.) HCI 2014. CCIS, vol. 435, pp. 496–500. Springer, Cham (2014). doi:10.1007/978-3-319-07854-0_86
Author information
Authors and Affiliations
Corresponding author
Editor information
Editors and Affiliations
Rights and permissions
Copyright information
© 2017 Springer International Publishing AG
About this paper
Cite this paper
Satoh, H., Shibata, K. (2017). Adaptation Monitoring System Preventing Fall Down from a Bed for Individual Difference of Behavior. In: Stephanidis, C. (eds) HCI International 2017 – Posters' Extended Abstracts. HCI 2017. Communications in Computer and Information Science, vol 714. Springer, Cham. https://doi.org/10.1007/978-3-319-58753-0_42
Download citation
DOI: https://doi.org/10.1007/978-3-319-58753-0_42
Published:
Publisher Name: Springer, Cham
Print ISBN: 978-3-319-58752-3
Online ISBN: 978-3-319-58753-0
eBook Packages: Computer ScienceComputer Science (R0)