Fall Detection by Helmet Orientation and Acceleration during Bike Riding Ho-Rim Choi 1, Mun-Ho Ryu 2,3, Yoon-Seok Yang 2, Nak-Bum Lee 4 and Deok-Ju Jang 5 1 Department of Healthcare Engineering, Chonbuk National University, 567 Baekje-daero, Deokjin-gu, Jeonju-si, Jeollabuk-do 561-756 Republic of Korea 2 Division of Biomedical Engineering, Chonbuk National University, 567 Baekje-daero, Deokjin-gu, Jeonju-si, Jeollabuk-do 561-756 Republic of Korea 3 Center for Healthcare Technology Development, Chonbuk National University, 567 Baekjedaero, Deokjin-gu, Jeonju-si, Jeollabuk-do 561-756 Republic of Korea 4 Technology Licensing Center, Industrial-Academic Cooperation Foundation, Chonbuk National University, 567 Baekje-daero, Deokjin-gu, Jeonju-si, Jeollabuk-do 561-756 Republic of Korea 5 eclio Co., Ltd., 761-10, 2ga, Palbok-dong, Deokjin-gu, Jeonju-si, Jeollabuk-do, Republic of Korea {ck-ccr, mhryu, ysyang, biomecca}@jbnu.ac.kr, jdjjang@cliotech.co.kr Abstract. This study represents research on falls while riding bicycles. We present an algorithm that detects the occurrence and direction of falls during bicycle riding through the use of a sensor module installed on the helmet. The sensor module is equipped with a tri-axial accelerometer, tri-axial gyroscope, and tri-axial magnetometer to record acceleration and tilt signals. We determine the threshold of the algorithm for fall detection based on experiments of riding over speed bumps and performing warm-up exercise motions prior to riding a bicycle. The algorithm we developed achieved 98% accuracy in fall detection and 98.5% accuracy in detection of fall direction. This study performed simulations indoors for the sake of subjects safety; however, we expect to produce the research in an environment more similar to real-life situations. In future, it is also possible that the developed system will be applied to an emergency reporting system for bicycle falls. Keywords: Fall Detection, Bike Riding, Inertial Sensor, Helmet. 1. Introduction Bicycles are an economical and eco-friendly means of transportation and are used widely for commuting. As the amount of spare time increases and interest in health intensifies, more people enjoy riding bicycles for leisure [1, 2]. When people fall during bicycle riding, they may sustain injuries ranging from bruises to bone fractures and, in the worst case, even brain injuries. For the elderly, the risk may be much higher. As a result, people must wear helmets when they ride bicycles. 59
There has already been considerable research conducted on the analysis of patterns of human motion in activities of daily living (ADL) and fall detection. Nyan et al. attached a tri-axial accelerometer to subjects clothing at the shoulder and detected lifestyle patterns in the time frequency domain [3]. Falls are detected using peak values of tri-axial acceleration signals and a system was set up to call for help on short notice through short message service (SMS) when falls occur. Hwang et al. developed an algorithm of fall detection using a tri-axial accelerometer, tri-axial gyroscope, and a tiltmeter [4]. If the average value of the tilt sensor exceeded the first threshold, the differential value of acceleration signals was calculated. If the differential value exceeded the second threshold, it activated a timer and a 10 s countdown was initiated. Because people move less frequently after taking a fall, the event is deemed a fall if the differential value of acceleration after 10 s fails to exceed the third threshold. Furthermore, Bourke et al. conducted research to distinguish ADL from falls by measuring the pitch and roll of the body using a bi-axial gyroscope [5]. They designed a specific algorithm based on the integration of angle and speed data, and each case of acceleration that differentiated each speed. However, while integrating the speed data, the problem of drift occurs in which offsets are integrated, due to which accurate angle would not be obtained. The abovementioned studies address falls in ADL but there has been no research on falls during bike rides. The difference between the risk of falls in ADL and that during a bike ride is significant. The velocity of a forward-moving bike compounds the velocity of a fall, resulting in a much stronger shock; therefore, the danger of a fall during a bike ride is greater. Consequently, research on fall detection during a bike ride is required. Thus, this study developed a system of fall detection during bike riding rather than falls during ADL. We designed an algorithm of fall detection by fixing on a leisure bike helmet a sensor module in which a tri-axial gyroscope, tri-axial accelerometer, and tri-axial magnetometer were installed. The algorithm of fall detection was applied by calculating the convergence of acceleration signals emitted by the sensor module and the vector signals of the rotation angles of quaternions. Along with fall detection, fall direction was identified by analyzing vector signals of rotation angle. The experiment proceeded with forward falls, backward falls, leftward falls, and rightward falls, with the subject sitting on the saddle of a bike indoors for the sake of the subject s safety. Moreover, additional experiments were performed with speed bumps and warm-up exercises prior to riding a bike, in order to set the threshold to be used for the algorithm. 2. Methods 2.1 Hardware system The sensor unit employed in the study consisted of a tri-axial gyroscope, a tri-axial accelerometer, and a tri-axial magnetometer, along with a 32-bit multipoint control 60
unit (MCU, EBIMU 9-DOF, E2BOX, Korea). All of these components were built in to the sensor unit. The sensitivity of the gyroscope (L3G4200D, STMicroelectronics) was ±2000 dps, that of the accelerometer (LSM303DLH, STMicroelectronics) was ±8 g, and that of the magnetometer (LSM303DLH, STMicroelectronics) was ±2.5 gauss. The sampling speed was 100 Hz. For wireless communication with the host PC, a Bluetooth module was used (Parani ESD-200, Sena, Korea) (Fig. 1). Fig. 1. Configuration of the sensor unit. 2.2. Experiment Protocol Five healthy male subjects (ages 20 28, 24.2 on average) participated in this experiment. The subjects fell to the front, back, left, and right directions from a seated position on a 60 cm high bicycle saddle, placed on a mattress indoors for safety, while wearing a helmet with the sensor module on the head. In this study, the subjects were induced to fall unintentionally and naturally by moving the mattress quickly using pneumatic pressure. All the experiments were repeated 10 times for each direction. 2.3. Algorithm of Fall Detection This study used signal vector magnitude (SVM), calculated as in Eq. (1), by applying a low-pass filter with a cutoff frequency of 5 Hz to tri-axial acceleration signals ( a x, a y, a z ) for fall detection. 2 2 2 a = ax + ay + az. (1) Because a can change due to the shock from traversing a speed bump and the acceleration added from riding the bicycle, a fall cannot be detected from a alone. Therefore, additional analysis is performed on the tilt of the sensor. The angle data used in this study were the rotation angle vectors ( θ x, θ y, θ z) of quaternions produced by the sensor unit, which were then converted from radians to degrees. Here, θ x, θ y, θ z are respectively equivalent to the rotations centering on X, Y, and Z. As the angle data on the vertical axis is unrelated with the fall, the data on vertical axis angles were ruled out when calculating the tilt ( θ tilt ) (Eq. (2)). 61
tilt 2 x 2 y θ = θ + θ. (2) The overall process of fall detection is as follows (Fig. 2). First, if the peak value of a does not exceed a th, the experimental trial is classified as a normal ride. If the peak value of a exceeds a th, the tilt of the sensor is analyzed. If the peak value of θ tilt does not exceed θ th, the trial is identified as a speed bump or an ADL motion, but if it exceeds θ th, it is identified as a fall during a bike ride. Fall direction is detected by comparing the variations ( θx, θ y after a fall has occurred. If θx is greater than θy) of θ x and θy and θ x is greater than 0, the trial is identified as a forward fall, and if θ x is smaller than 0, it is identified as a backward fall. Likewise, if θy is greater than θx and θ y is greater than 0, it is identified as a leftward fall, and if θ y is smaller than 0, it is identified as a rightward fall. Fig. 2. Flow chart of the fall detection algorithm. 3. RESULTS In this study, we designed an algorithm to detect the occurrence and direction of falls based on the parameters a and θ tilt. Parameter measurements were obtained from the convergence of readings generated by a sensor module with a built-in triaxial accelerometer, tri-axial gyroscope, and tri-axial magnetometer, installed on a helmet worn by the experimental subjects. When the algorithm of fall detection was applied, falls were not detected in the preliminary experiments involving riding on a 62
flat surface, traversing a speed bump, and performing warm-up exercises. When riding on the flat surface, no fall was detected because a and θ tilt did not exceed a th and θ th (Fig. 3(a)), and when traversing the speed bump, no fall was detected because θ tilt did not exceed θ th, though a exceededa th (Fig. 3(b)). In addition, during warm-up exercises consisting of bending the upper body forward, no fall was detected because θ tilt exceeded θ th but a did not exceed a th (Fig. 3(c)). When a fall occurred during a bike ride, however, the fall was detected because a and θ tilt exceeded a th and θ th, respectively (Fig. 3(d)). Fig. 3. (a) a (top) and θ tilt (bottom) when riding on a flat surface, (b) a (top) and θ tilt (bottom) when riding over a speed bump, (c) a (top) and θ tilt (bottom) when performing warm-up exercises (bending the upper body forward), and (d) a (top) and θ tilt (bottom) when a fall occurs. Five experimental subjects executed a total of 200 falls comprising 10 trials in each direction, classified as front, back, left, and right, starting from a seated position on a bicycle saddle, in order to verify the accuracy of detection of falls and fall direction. As a result of applying the developed algorithm, the fall detection rate recorded was 98%. The algorithm also demonstrated detection capability of 98.5% for fall direction (Table 1). 63
Table 1. Accuracy of fall detection and fall direction detection. Attempts Detections Total (%) Fall detection 200 196 98 Direction detection 200 197 98.5 4. CONCLUSIONS In this study, we developed a system to detect falls during bike rides while wearing a leisure helmet on the head. We developed an algorithm to detect the occurrence and direction of falls using signals of acceleration and rotation angle vector of quaternions. Parameter readings are emitted by a sensor installed on the leisure helmet, containing a sensor module with a built-in tri-axial accelerometer, tri-axial gyroscope, and triaxial magnetometer. Fall detection was achieved by judging the acceleration signal and the rotation angle vector, and the detection of fall direction was determined based on the rotation angle vector. When fall and fall direction were detected during a bike ride by applying the developed algorithm, it demonstrated excellent detection capabilities for both. This study was performed indoors rather than on actual bike rides outdoors. Thus, there is a need to conduct further experiments in environments more similar to real bicycle-riding conditions. In future, the sensor installed on the helmet can be applied to a system in which the sensor reports an emergency by transmitting emergency signals to mobile devices when a fall occurs during a bike ride. Acknowledgments: This work was supported by the Basic Science Research Program through the National Research Foundation of Korea (NRF) funded by the Ministry of Education, Science and Technology (2011-0010715). References 1. Gatersleben, B., Appleton, K.M.: Contemplating Cycling to Work: Attitudes and Perceptions in Different Stages of Change. Transportation Research Part A 41, 302--312 (2007) 2. Heinen, E., Maat, K., Wee, B.V.: The Role of Attitudes toward Characteristics of Bicycle Commuting on the Choice to Cycle to Work over Various Distances. Transportation Research Part D 16, 102--109 (2011) 3. Nyan, M.N., Tay, F.E.H., Manimaran, M., Seah, K.H.W.: Garment-based Detection of Falls and Activities of Daily Living using 3-axis MEMS Accelerometer. J. Phys.: Conf. Ser. 34, 1059--1067 (2006) 4. Hwang, J.Y., Kang, J.M, Jang, Y.W., Kim, H.C.: Development of Novel Algorithm and Real-time Monitoring Ambulatory System Using Bluetooth Module for Fall Detection in the Elderly. In: Proceeding of the 26th Annual International Conference of the IEEE EMBS, pp. 2204--2207. IEEE Press, San Francisco (2004) 5. Bourke, A.K., Lyons, G.M.: A Threshold-based Fall-detection Algorithm Using a Bi-axial Gyroscope Sensor. Med. Eng. Phys. 30, 84--90 (2008) 64