INTRODUCTION

Scientific gait analysis is the exploration of sensor patterns while walking. Gait analysis is primarily carried out in one of two ways: in a motion laboratory, with full analysis of the motion of body segments using highly accurate computer based force and optical tracking sensors, or in an office with the clinician making visual observations. The first method is expensive, the second method is inexpensive but requires substantial time and clinical expertise.

Falls in the elderly are very common occurrences that can have dramatic health consequences. According to the United Nations Population Division statistics, by the end of 2009 the elderly population reached 737 million, accounting for 10.8% of the total population. In the year 2025 it is projected to account for 15% of the total population. For people of 70-75 years old, the estimated incidence of falls is over 30 percent per year [1]. Nearly half of nursing home patients fall each year, with 40 percent falling more than once [2, 12]. In order to predict the risk of falls clinical teams are often put in situations of making treatment assessments without much in the way of solid data. They may see a patient after long time, relying on memory and subjective descriptions of progress. This high risk of falling among the growing elderly population has greatly influenced research on fall detection and fall prediction; the latter, being very challenging since to prevent a fall, we first need to identify the patterns correlated with a fall for a specific person.

The cost of Smartphones has decreased and their computational capabilities have rapidly increased with improvements in mobile technology. As self-contained devices, Smartphones present a mature hardware and software environment for developing various fall detection as well as activity monitoring systems. A Smartphone-based falls observation system can function almost everywhere, since mobile phones are highly portable. Currently, most Smartphones have sensors to observe acceleration, location, orientation, ambient lighting, sound, imagery, etc. [3]. These integrated sensors along with the pressure sensor shoes strengthen the capabilities of the Smartshoe. Smartshoe can already automatically detect falls. Indeed other teams have already created some Smartphone-based fall detection systems [4].

This fall prediction system leverages Smartphones as a programmable platform for monitoring well-being as people go about their lives [5]. The smartphone can infer a range of behaviors on the phone in real-time. Someday this could allow users to receive real-time feedback on their movement. In addition, with the improvements in mobile technology, the popularity of smartphone app stores can make any new technology instantly widely available.

This paper concentrates on the development of a pair of shoes for user movement identification for predicting a fall. Foot pressure signals can identify behavior of human gait and posture as reflected in foot pressure distribution. Many studies describe foot pressure as a detection system, but few have used smartphone and a smartshoe for the analysis. We report on a new smartphone and smartshoe used to predict a fall. Our major contributions are as follows:

• Proposed a smartphone- and smartshoe-based system to identify the quality of step activity in any environment.
• Developed a Wi-Fi network (my-adhoc) for the communication between smartphone and smartshoe.
• Provide users, health care professionals and caregivers with highly personalized health feedback.

Our system targets fall prediction among people with impairments that affect gait and balance, predisposing individuals to falling. These include common rehabilitation diagnostic groups and elderly populations, but also may eventually help identify gait disorders among children and be helpful in environment monitoring.

RELATED WORK

Veltink, et al. sensors are placed at toe or heel to recognize movements by thresholds. [6]. R.C. Luo, et al. explained adopting the methodology of information cognition from multisensory was regarded not only efficient but also reliable [7]. For diabetics, Morley et al. [8] and Maluf et al. [9] developed an insole-based system to quantify the conditions inside the shoe. Pappas et al. [10] used a pattern recognition algorithm to define the changes during the gait cycle using their device comprising of three force-sensitive resistors (FSR) located on an insole (one under the heel, and two at the first and fourth metatarsal heads), and a gyroscope. The system was tested on two subjects with incomplete spinal injury and was used to trigger functional electrical stimulation (FES), with demonstrated benefit for both subjects. In [11], the author proposes a method that uses a network of fixed motes to provide location information about the victim after a fall has been detected. iFall [4] is an Android application that has been developed to detect fall events in the fall detection territory. Data from the accelerometer is evaluated using several threshold-based algorithms and position data to detect a fall.

To address the drawbacks of the above-mentioned systems, we developed a smartphone- and smartshoe based system for human movement identification to predict falls. Moreover our design is highly secure and inexpensive because it requires only a smartphone with low cost smartshoes.

SYSTEM DESCRIPTION

Primary features of the foot pressure sensing shoes are its unobtrusiveness and portability. The wearable nature of shoes allows it to collect user’s motion signal freely. The schematic of the pressure sensing system is presented in Figure 1.

For the analysis of the kinematic motion of the foot, four piezoresistive pressure sensors were placed at the bottom of the shoe to assess the timing parameter and pressure distribution. A human foot usually divide by three different region, Fore Foot (FF), Mid Foot (MF) and Rear foot (RF). Most of the body pressure is measured from the rear foot and from the fore foot. Considering these issues we have placed two of our sensors in the fore foot region and two of them are in the rear foot region as shown in Figure 2.  We have used the flexiforce piezoresistive [13] force sensor for measuring the pressure while walking. The resistance of this sensor changes with the change in pressure.  The harder you press, the lower the sensor’s resistance. Resistance changes only when pressure is applied to the round area at the end of the sensor, but the resistance does not change while being flexed.

The SmartShoe is comprised of four piezoresistive pressure sensors, one arduino, and one wifly shield [13] with a battery power supply. We were using arduino, wifly shield and amplifier circuit as Wi-Fi communication module on shoe. This module would amplify the signal and convert it into digital signal. This data is transferred to the Smartphone through a my_adhoc Wi-Fi communication network.

In order to process this data, the communication module has two different software tasks. One is for the arduino and another is for the iPhone. In arduino, we programmed to read an analog signal from the shoe sensors and create a data packet to convert the signal into digital form Pressure data was collected for the people over period of time and every time a subject was tested with standing, walking, going upstairs, and going downstairs.  

Figure 3 portrays an early prototype with embedded sensors attached to the computational board, wifi and power components.

Volunteers were recruited for the validation of the smartshoe system. The testing involved placing the smartshoe instrumentation on the subjects’ own walking shoes. Each subject was asked to perform a series of locomotor tasks, while systems simultaneously collected data.

A total of 5 subjects were recruited. Characteristics for each group are summarized with means (and standard deviations) in Table 1.

Each subject first walked at his or her own self-selected natural pace for 2 to 4 trials, termed “free gait.”

PRINCIPLES OF THE SYSTEM

In the system, the smartshoe is used to collect the foot pressure value while the subject would asked to perform four different types of movement, stand still, Walking,  going upstairs, and going downstairs. After receiving the data through Wi-Fi communication we have processed it inside the mobile phone to identify the movement steps for predicting fall. At that moment, the system detects a high-risk gait pattern and enables a warning to the subjects through an audio message and vibration, to alert them about an imminent fall.

Human walking is a cyclical movement, so here we use the similarity of the data between two adjacent cycles to assess the walking stability.

Stability = Similarity (Ci, Ci+1) ----------(1)

Where Ci is the data of the preceding cycle, and Ci+1 is the data of the next cycle.

The dynamic symmetry of gait is defined as the discrepancy of bilateral data in gait cycle on all symmetric attributes.

Symmetry = Discrepancy (Ri, Li) --------(2)

Where Ri is the data of the right-side and Li is the data of the left-side on one attribute.

The average accuracy with four sensors in each shoe is 85.5% for all four kind of movements and the identification accuracy for each moving pattern is in table 2.

CONCLUSIONS

In this paper we presented a smartphone- and smartshoe-based human moving pattern recognition system for predicting falls. This design was observed to collect discriminatable data. More data are needed to increase the fall prediction accuracy for a wider variety of people and better target the intended population. This work, however provides a solid conceptual and demonstrable foundation for the next phase of design.