You are on page 1of 3

Dynamic Fall Detection and Pace Measurement in Walking Sticks

Oscar Almeida, Ming Zhang, Jyh-Charn Liu Dept. of Computer Science, Texas A&M University {oscar10, zhangming, jcliu} @tamu.edu Abstract
Falls are increasingly among the leading causes of elderly injuries and deaths each year. Several of these victims depend on a walking stick or cane for support while walking. Rendering aid more quickly to those who fall may decrease the severity of injury in several cases. In this paper, we propose a dynamic fall detection system embedded into walking sticks and canes. By using a gyroscope chip to measure angular velocity of the stick, we can detect when a user may have fallen. Also monitored by the detection system is the users walking pace, such that the user will be warned when traveling at paces above his or her normal speed. With different polling frequency levels to preserve energy, we present a low-power device that can potentially improve safety among the elderly. experienced by more than half of Americans over the age of 80 [4]. Lastly, a lack of physical exercise becomes a risk factor, especially when one of the previous risk factors is also met [3]. Due to the high frequency and severity of falls in the elderly population, it is vital that a fall be detected and reported immediately. In this paper, we propose that embedding a dynamic fall detection system into walking sticks and canes will significantly reduce the time necessary to render aid to a fall victim. Section 2 describes the survey of technology and high-level system description. Following is the signal processing algorithm for the gyro output, leading to our methodology behind the proposed fall detection. Section 5 illustrates the userspecific pace measurement benefits of this device, and section 6 describes the power considerations taken.

2. System Overview 1. Introduction


Historically, walking sticks have provided a means of wardrobe accessories, protection, and aid to the balanceimpaired. Today, walking sticks are overwhelmingly used by the elderly who need assistance in walking. By holding a stick or cane either in the dominant hand, or in the hand opposite the weakness or injury, the user can shift his or her weight away from the weaker side of the body. According to the National Center for Injury Prevention and Control, more than one-third of adults over the age of 65 fall each year. In addition, falls are increasingly the leading cause of injury deaths and nonfatal injuries of the elderly. Falling down can result in bone fractures and or brain trauma [1]. Several risk factors increase the probability of an elderly person becoming a victim of a fall. Osteoporosis, a disease that causes bones to become more fragile, allows bones to be more susceptible to fractures. In fact, osteoporosis can progress through the body painlessly until such a fracture occurs [2]. Next, impaired vision and environmental hazards can cooperatively cause harm to the unaware walker. At least one-third of all falls in the elderly involve environmental hazards in the home [3]. Another risk factor, glaucoma, results in a loss of vision by damaging the optical nerves. About half of the people in the United States that have glaucoma do not know it. Another type of visual impairment is a cataract. These are Our proposed dynamic fall detection device consists of two main parts: the fall detection device and the alarming device. When a fall is detected, the user is asked to push a reset the stick via a switch. If this action is not carried out in a timely manner, the alarming device will become activated. This paper focuses only on the fall detection itself, while the alarming procedure is left as a future area of research. The underlying concept behind dynamic fall detection is that while the walking stick is in use, the sticks angular velocity is constantly being polled. Consider a static fall detection system based on orientation alone [5]. A false negative may occur if a victim falls but the stick never reaches a horizontal position. Also, a false positive will occur every time the stick rests unused in a horizontal position. Our proposed dynamic methodology for fall detection is more likely to see the data signal cross some angular velocity threshold during any fall, even if the stick does not fall completely to the floor. The second objective of our design is to monitor the users walking pace to identify abnormalities. Our testing prototype is driven by an Atmel EB63 microprocessor evaluation board and a Gyration MG1101 MicroGyro. The gyroscope chip (gyro), which measures angular velocity, is polled at an appropriate frequency to detect if the user may have fallen. The EB63 must poll the gyro for input data, process the data, and when

triggered, signal for help in the event that a fall is detected. The board is also responsible for power management. Because our algorithms require only a small amount of calculations, a small processor chip would be capable of performing the required operations.

3. Signal Processing
A noisy gyro output signal can occur due to wiring, along with vibrations of the gyro as the stick makes contact with the ground. Figure 1(a) shows an example of raw data output from the gyro. Point A represents the first of two parts of a step with a walking stick. This includes the user placing the sticks base on the ground in front of him or herself. Point B represents the step taken by the user to stand beside the stick. We observe that the local minima and maxima are not very well defined in this signal, and attribute the cause to noise. In order to eliminate minor signaling errors, we propose the use of a small, weighted running average, low-pass filter. While this filter is targeted specifically at local minima and maxima, the integrity of the global maxima and minima are preserved. This smoother processed output signal is more representative as a whole of the walking sticks motion than the raw data. Figure 1(b) shows the same data after being processed by the running average low-pass minima/maxima filter. This smoother representation of the gyro output is used for fall detection.

Figure 2: Walking stick prototype with gyro and Atmel EB63 evaluation board. Orientation of the gyro, located at the base of the stick, is displayed on the left.

4. Fall Detection
Because its main function is to generate vertical support to the user, a canes angular range of motion is extremely limited. We define T to be the maximum angle from vertical that is achievable during normal operation. Additionally, we assume the user is stable after each step successfully taken. Movement away from a stable point represents the walking sticks stability. Mathematically, we define stability as a summation of sequential gyro outputs, such that each output is greater than a given threshold value Treset. Another reason the stability is reset with every step is that the gyro does not provide the system with a point of reference in terms of displacement. It is impossible to maintain a net sum of angular velocities accurately enough to determine the exact net displacement of the stick. Displacement would also be dependent on the frequency at which the gyro is polled. The most current readings of the angular velocities in axes A and B are represented by At and Bt respectively. Calibration of the stick occurs by creating a physical mapping of T to a data threshold Tfall. Another data element, R, as shown in Eq. 1 represents the current magnitude of the resultant angular velocity from axes A and B combined. Eq. 2 then uses the value of R to define the stability S. If S is greater than or equal to Tfall at any point in time, then it is assumed that the user may have fallen, and the alarming device should be activated.

(a)

(b) Figure 1: (a) Raw and (b) filtered gyro data.

R = At2 + Bt2
R < Treset 0, S = S + R, R Treset

Eq. 1 Eq. 2

5. User-specific Design
In addition to fall detection, we present a method of characterizing users based on their walking pace. If a user is determined to be walking above their normal average pace, the stick can warn the user to slow down. Angular velocity data is polled at a sampling rate of 15 Hz. Each velocity datum has two orthogonal components, along axes A and B. One axial component may be significantly larger than the other. In such a case, we consider only the axis with a larger magnitude for the pace calculation. The first 50 data points taken are used to make the decision to choose the axis to be used. A moving average over five data points is used to form a smoother data curve. Let V(k) denote the angular velocity at a point in time, k. We consider the time period between two adjacent peaks of angular velocity as one step. To find peaks, we use a moving average over 100 data points as the threshold. A search procedure is initiated to find the peak point from another point with a value larger than the threshold. The procedure is terminated if the next points value drops below the threshold. This peak point is marked with a vertical black line, as shown in Figure 3. To get the pace between two adjacent peaks, we first calculate the distance and then divide by the time interval (Eq. 3). Here, p1 and p2 represent the two adjacent peaks. A moving average over N steps is calculated and used to determine how fast the user is walking.
v=
p2 1 V (k ) p2 p1 + 1 k = p1

1 0.8 0.6 0.4 Angular Velocity 0.2 0 -0.2 -0.4 -0.6 -0.8 -1

x 10

20

40

60

80

100 120 Data point

140

160

180

200

Figure 3

6. Low-power Design
In order to maintain the walking stick as a low-power device, we implement different modes of operation. Each mode corresponds to a unique polling frequency. For normal operation, we chose a low sampling rate of 15Hz. If the stick is assumed to be idle, we reduce the polling to 1Hz. This frequency is still enough to detect a step. If the stick remains idle, the stick enters a deep sleep mode, polling only once every ten seconds until motion is detected. Of course, signal processing will be disabled while in either sleep mode.

Eq. 3

7. References
[1] Centers for Disease Control and Prevention, Falls Among Older Adults: An Overview, Center for Disease Control and Prevention, 2007. [Online]. Available: http://www.cdc.gov/ncipc/factsheets/adultfalls.htm. [2] National Osteoporosis Foundation, Osteoporosis: A debilitating disease that can be prevented and treated, National Osteoporosis Foundation, 2007. [Online]. Available: http://www.nof.org/osteoporosis/index.htm. [3] K. R. Tremblay Jr. and C. E. Barber, Preventing Falls in the Elderly, Colorado State Univerity Cooperative Extension, 2006. [Online]. Available: http://www.ext. colostate.edu/PUBS/CONSUMER/10242.html. [4] National Eye Institute, Cataract, National Eye Institute, 2006. [Online]. Available: http://www.nei.nih.gov/health/cataract/cataract_facts.asp [5] MedGadget, i-Stick, and Intelligent Walking Stick, MedGadget, 2006. [Online]. Available: http://medgadget. com/archives/2006/11/istick_help_ive.html

We performed four experiments, walking at different paces. Two were slower, while two were at a faster pace. The summary of these experiments are show in Table 1. One can see that there is a large difference between a slow and fast walking pace. The normal pace can be achieved by a simple calibration procedure. Table 1 Walking #Data Exp. ID Avg. pace speed points 1 Very Slow 2736 401 2 Slow 1926 551 3 Fast 1049* 751 4 Very Fast 1246 1200 (* distance walked was 25% less than in Exp. 1, 2 and 4)

You might also like