Detecting falls using a wireless sensor network

3:35 pm - 4:00 pm 23 Wednesday

Track

Health & Biomedicine

 

INTRODUCTION

The aging of the world population has lead to new health problems. People now live longer, but unfortunately the quality of life of the elderly is deplorable in many cases. The number of old people affected by brain disorders have increased significantly in recent years. Among these disorders, dementia, which has no cure, is one of the major health problems of the elderly.
 
Wandering and falls, which are closely related, are among the problems that most significantly affect patients with dementia. Wandering may lead to falls, and consequently, tosevere injuries. In fact, nearly one third of people over 75 years old suffer at least one fall every year. Also, falls are one of the main causes of death or severe injuries of people over 65 years old. This is the reason why people with dementia need constant supervision, and impose a heavy burden to families and caregivers. To alleviate this problem, the development of fall detection systems for the elderly is an important research topic[1].
 
In this paper, a non-invasive falls detection system for the elderly, based on the use of WSN, is proposed. The main objective of the system is to detect in real-time if someone has fallen, and to alert the caregivers to provide assistance. A rapid response may help to provide timely medical assistance, and prevent as much damage as possible.
 

METHODOLOGY

The proposed system uses the acoustic signal sensed by the motes, as well as signal processing and pattern recognition techniques, to detect a fall. To illustrate this, Fig. 1 shows the block diagram for the proposed system.
 
Data sampling. The nodes of the WSN, placed at fixed locations (i.e., the room of the patient), are equipped with a microphone, and are constantly sensing the environment. When the intensity of the collected sound is larger than a threshold, the gathered data are sent to the sink. The sink uses these data as input of a signal-processing algorithm based on the use of cross-correlation. Since sampling rate is an important factor in pattern recognition, based on experimental results and to the to Nyquist theorem[2], a fixed sampling rate of 4 KHz was used.Cross-correlation (alignment). To avoid any missed event, sequences are aligned using a template as a reference, which was obtained using a real fall event. The cross-correlation of two time sequences measures the similarity between a sequence, and shifted similar sequences as a function of the lag. Features extraction. The MFCC have been widely used for speech/audio recognition [3], and represents the short-term power spectrum of a sound, based on a linear cosine transform of a log power spectrum on a nonlinear Mel scale of frequency. The MFCCs take into consideration human perception sensitivity with respect to frequencies.
 
Initially, the sampled data are stored in small frames of N samples, where the most used value for N is 256. The frames then must be converted from time domain to frequency domain, this is archived by the use of fast Fourier transform (FFT). A Mel filters bank will process the frequency domain data. In order to create the filter bank on Mel scale, a frequency conversion is needed. Eqn. 1 is used to compute the Mel scale convention, where M(f) is the frequency on Mel scale. A lower and an upper frequency cutoff are needed, where the upper frequency on the filter bank is limited to half the sampling rate frequency [4].
 
This is the process to convert the log Mel spectrum into time domain using discrete cosine transform (DCT). The result of the conversion is called Mel frequency cepstrum coefficient. The set of coefficient is called the acoustic vectors. Therefore, each input utterance is transformed into a sequence of acoustic vector. The output, after applying the DCT, is known as Mel frequency cepstrum coefficient.
 
Matching process. The DTW algorithm measures the similarity between two temporal sequences, and it is also used for speech recognition [5]. The DTW algorithm can calculate the optimal alienation path from two series or templates, representing the measured similarity with a coefficient called the total distance cost [6].The absolute distance between the values of twosequences is calculated using the Euclidean distance computation. Each matrix element (i, j) corresponds to the alignment between i and j. Eqn.2calculates the accumulated measured distance.
 
Finally, when the calculated total distance cost is less or equal than a given threshold, the system concludes that a fall has occurred. Otherwise, the system ignores the event. It is important to note that this threshold is obtained for the particular scenario where the system is used, considering the patient, the floor of the room, the ambient noise, among others.
 

CONCLUSIONS

Many health problems have increased as the world’s population ages. Among them, accidental falls are one of the main causes of injuries of the elderly. In this paper, a non-invasive falls detection system, based on the use of WSN, was proposed. If a fall is detected, an alert signal is sent to caregiver in order to provide timely medical assistance, and prevent as much damage as possible.
 
To validate the proposed system, we built a prototype and conducted a series of experiments. The obtained results from the experimental tests showed that the proposed system has a detection rate of 90%when no acoustic interference is present. Otherwise, the detection rate is nearly 80%.However, it can be improved adding more sensors, which may allow the detection of other important causes of injuries, such as wandering.
 

REFERENCES

[1]N. Batsch and M. Mittelman, "World Alzheimer Report 2012: Overcoming the stigma of dementia," London: Alzheimer’s Disease International, vol. 75, 2012.
[2]R. K. Bansal, A. Goel, and M. K. Sharma, MATLAB and Its Applications in Engineering: Pearson Education, 2009.
[3]C. K. On, P. M. Pandiyan, S. Yaacob, and A. Saudi, "Mel-frequency cepstral coefficient analysis in speech recognition," in Computing & Informatics, 2006. ICOCI'06. International Conference on, 2006, pp. 1-5.
[4]S. Young, The HTK Book: Entropic Cambridge Research Laboratory, 1997.
[5]C. S. Myers, A Comparative Study of Several Dynamic Time Warping Algorithms for Speech Recognition: Massachusetts Institute of Technology, Department of Electrical Engineering and Computer Science, 1980.
[6]"Dynamic Time Warping," in Information Retrieval for Music and Motion, ed: Springer Berlin Heidelberg, 2007, pp. 69-84.
Speaker:
Edgar Dominguez