Fall detection system based on BiLSTM neural network
DOI:
https://doi.org/10.14738/tmlai.75.6318Abstract
The purpose of this article is to analyze the characteristics of human fall behavior to design a fall detection system. The existing fall detection algorithms have problems such as poor adaptability, single function and difficulty in processing large data and strong randomness. Therefore, a long-term and short-term memory recurrent neural network is used to improve the effect of falling behavior detection by exploring the internal correlation between sensor data. Firstly, the serialization representation method of sensor data, training data and detection input data is designed. The BiLSTM network has the characteristics of strong ability to sequence modeling and it is used to reduce the dimension of the data required by the fall detection model. then, the BiLSTM training algorithm for fall detection and the BiLSTM-based fall detection algorithm convert the fall detection into the classification problem of the input sequence; finally, the BiLSTM-based fall detection system was implemented on the TensorFlow platform. The detection and analysis of system were carried out using a bionic experiment data set which mimics a fall. The experimental results verify that the system can effectively improve the accuracy of fall detection to 90.47%. At the same time, it can effectively detect the behavior of Near-falling, and help to take corresponding protective measures.
References
(1) Mengyu Wu, Weihua Yu, Qian Ge, Cai Xuan, Jinping Qian, Ziwei Wang. Meta-analysis of the effects of exercise intervention on the fall, gait and balance of debilitated elderly [J]. Evidence-based nursing, 2018, 4 ( 11): 966-972 (in Chinese)
(2) Yu Zheng. Design and implementation of community elderly fall monitoring system based on ZigBee and Ethernet [D]. South China University of Technology, 2018. (in Chinese)
(3) Yuan Jie. Vision-based design and research of fall detection system for the elderly [D]. Jiangxi University of Technology, 2018. (in Chinese)
(4) Ding Yafei. Design of Fall Detection and Remote Monitoring System Based on Wearable Devices[J]. Information and Computer (Theoretical Edition), 2018(04) (in Chinese)
(5) Shi Xin. Research on motion human body behavior recognition based on pressure-sensing gait[D]. Chongqing University, 2010. (in Chinese)
(6) Niu Dezhen, Liu Yawen, Cai Tao, Peng Changsheng, Zhan Yongzhao, Liang Jun. A fall detection system based on recurrent neural network[J]. Journal of Intelligent Systems, 2018, 13(03): 380-387. (in Chinese)
(7) FANG-YIE LEU, CHIA-YIN KO, YI-CHEN LIN, et al.Smart Sensors Networks[M]. United Kingdom: Mara Conner, 2017: 205–237
(8) CHEN L, MA H T, LIU S, et al. Posture estimation by Bayesian Network with Belief Propagation[C]//TENCON 2013-2013 IEEE Region 10 Conference. Xi’an, China,2013: 1–4.
(9) Duan K B, Keerthi S S. Which is the best multiclass SVM method? An empirical study[M]// Multiple Classifier Systems. Springer Berlin Heidelberg, 2005:278–285.
(10) Zhai Yujian,Hua Liang,Chen Ling,Gu Juping,Shen Wei.Study on Intelligent Recognition of Falling Old Man Based on Multi-weight Neural Network[J].Science and Technology and Engineering,2015,15(04).(in Chinese)
(11) SZ Erdogan, TT Bilgin, J Cho. Fall-down detection by using K-nearest neighbor algorithm on WSN data[C]//GLOBECOM Workshops. Houston, USA, 2011:2054–2058.
(12) Vaidehi V, Ganapathy K, Mohan K, et al. Video based automatic fall detection in indoor environment [C].In: Proceeding of International Conference on Recent Trends in Information Technology, 2011:1016-1020.
(13) Kwolek B , Kepski M . Improving fall detection by the use of depth sensor and accelerometer[M]. Elsevier Science Publishers B. V. 2015.
(14) LUO D, LUO H, SCHOOL I. Fall-down detection algorithm based on random forest[J]. Journal of computer applications, 2015, 35(11): 3157–3160.
(15) Shi Xin. Research on motion human body behavior recognition based on pressure-sensing gait[D]. Chongqing University, 2010. (in Chinese)
(16) HOCHREITER S, SEHMIDHUBERJ. Long short-term memory[J]. Neural Computation, 1997, 9(8): 1735-1780.
GRAVES A. Supervised sequence labelling with recurrent neural networks[D]. Ph. D Dissertation. Manno. Switzerland: Technical University of Munich, 2008.
