//Logo Image
Author: Che-Chang Yang(2006-09-01); recommendation: Yeh-Liang Hsu (2006-09-01).
Note: This article is Chapter 2 of Che-Chang Yang’s Master thesis “Development of a Portable System for Physical Activity Assessment in a Home Environment.”

Chapter 2. Accelerometry for human body movement

2.1 Motion sensors

For physical activity assessment, body-fixed motion sensors have been accepted as the preferred alternative to optical or magnetic sensors. Various motion sensors have been designed for this purpose, ranging from mechanical pedometers and actometers to electronic gyroscope and accelerometers.

2.1.1 Mechanical sensors

A pedometer, or a step-counter, is one of the earliest mechanical motion sensors (Figure 2.1). It is usually worn at the waist or at the ankle to count each step when jumping, walking or running. The impulse of each step when landing is registered in a counting mechanism. Although electrical pedometers are available, individual calibrations of a stride length are still necessary. Moreover, the pedometer cannot reflect and evaluate the intensity of movement and energy expenditure. Actometers that have been designed to record physical activity are not always the motion sensors in the metrological sense. They may differ largely from one another to meet specific applications. Schulman and Reisman [1959] proposed a modified actometer which was designed as a wristwatch which provides measurement correlating with energy expenditure.


Figure 2.1 mechanical pedometers

2.1.2 Electronic sensors

Electronic motion sensors, such as gyroscopes and accelerometers, are most commonly adopted in place of mechanical motion sensors due to the superiority of precise response to both frequency and intensity of body movement. Gyroscopes [Figure 2.2] are widely used in vehicle navigation. A gyroscope which consists of a vibrating element coupled to a sensing element is used to detect angular velocity. The operation principle of a gyroscope is based on the Coriolis effect which is an apparent force that arises in a rotational reference frame and is proportional to the angular rate of rotation. The advantage of using gyroscopes is that the angular rate signal is less noisy than an acceleration signal since acceleration is the derivative of velocity and involves higher frequency components. Yet, signal drift and being sensitive to shock and impulse are the major drawbacks of using a gyroscope [Najafi, 2002]. Aminian and Najafi used a gyroscope as the kinematic sensor to measure trunk tilt for physical activity assessment.

Figure 2.2 A gyroscope [Murata, Japan]

The use of accelerometers to measure body movements dates back to the early 1970’s [Moris, 1973]. Operating principles of electronic accelerometers are based on piezo-electric, piezo-resistive and capacitive properties. The piezo-electric element which behaves as a damped-spring generates an electrical charge in response to external forces. In a piezo-resistive accelerometer, the actuating element is replaced by a resistive silicon circuit that generates differential impedance in response to external forces. A capacitive accelerometer functions on the principle of differential capacitance. Acceleration causes displacement of a silicon structure and results in a change in capacitance. Change in capacitance is transduced into an analogue voltage which is proportional to acceleration. Piezo-resistive or capacitive accelerometers are usually smaller than piezo-electric ones. An important distinction is that piezo-resistive or capacitive accelerometers respond to constant acceleration as well as AC components while the piezo-electric accelerometers do not. Piezo-resistive or capacitive accelerometers are preferable due to the DC-responsive property which enables easier calibration. Most important of all, inclination angles of a body segment with respect to its fixed joint axes at rest can be determined by the response to the constant gravitational component. With MEMS manufacturing technology, tri-axial accelerometers can be miniaturized in small packages as shown in Figure 2.3.

Figure 2.3 Electronic accelerometers

In addition to those motion sensors, the goniometer, electronic compass and foot pressure pad have also been designed for this purpose. A goniometer is composed of a series of strain gauges in the flexible tube which connects two end blocks. It can be used to measure the angle between two segments as shown in Figure 2.4. An electronic compass uses magnetic effect to measure the orientation of a subject. Kemp et al. [1998] have proposed the possibilities of using an electronic compass to determine a body’s position. Foot pressure pads (or resistive switch) attached to the sole are used to analyze gait during locomotion [Zhu et al., 1991].

Figure 2.4 Goniometer [MindWare Technologies]

2.2 Introduction to Kionix KXM52-1050 tri-axial accelerometer

2.2.1 Kionix KXM52-1050 tri-axial accelerometer

In this research, the motion sensor used was the KXM52-1050 tri-axial accelerometer.  It was chosen as it has several advantages over other accelerometers. The Kionix KXM52-1050 tri-axial accelerometer, as shown in Figure 2.5, is a high performance silicon micro-machined linear accelerometer consisting of a sensing element and a CMOS signal conditioning Application Specific Integrated Circuit (ASIC) packaged in a standard 5×5×1.8mm DFN(DualFlat Non-lead). The device functions on the principle of differential capacitance. Acceleration causes displacement of a silicon structure resulting in a change in capacitance. A signal-conditioning CMOS technology ASIC detects and transforms changes in capacitance into an analogue voltage which is proportional to acceleration. The general specification of Kionix KXM52-1050 is listed in the Appendix.

Figure 2.5 Kionix KXM52 series accelerometers

Figure 2.6 shows the functional diagram of KXM52. Three sensors measure the accelerations along X, Y, and Z axes. The charge amplifiers transfer the differential capacitance (charge) into an analog signal and then amplify the signal to a higher voltage level. The oscillator generates differential charge and results in specific outputs when the self-test mode is activated. The built-in 32kW resistors and the external, user-defined bypass capacitors comprise low-pass filters for output signals.

Figure 2.6 The functional diagram of Kionix KXM52

2.2.2 Typical schematic

Figure 2.7 shows the typical schematic that is used in the postural sensory module. The output will change with supply voltage(Vdd) variations ranging from 2.7(or 2.5)V to 5.5V. 3.3V is recommended as the output of the KXM52-1050 is programmed and tested at 3.3V power supply. In this schematic, a 0.1mF capacitor (C1) cross Vdd to ground is recommended to decouple the noise of the power supply. For more decoupling, a ferrite bead or small resistors may be inserted into the supply line.

Figure 2.7 Typical schematic of KXM52

Capacitors C2 to C4 that cross the output pins to ground are used to adjust the output bandwidth of each axis. These capacitors with the 32kW internal resistor comprise RC low-pass filtering circuits. The desired bandwidth  is determined by the Equation (1):


The output bandwidth is an important parameter when using the accelerometer. For KXM52-1050, the output bandwidth of the X and Y axes ranges from DC to 3kHz, and that of the Z axis is from DC to 1.5kHz. The choice of desired bandwidth depends upon the application. For example, a higher output bandwidth is advisable for the applications which require fast and sensitive motion detection. As for human postural detection required in this research, only low bandwidth is needed. In Figure 2.7, 0.1mF capacitors are adopted, which limits the bandwidth at about 50Hz (Equation  (2)).


Pin 9 of KXM52-1050 is the power shutdown pin. When this pin is left floating or grounded, the KXM52 is shut down and draws very little power. When tied to Vdd, the unit is fully functional. Pin 10 is the self-test pin, and 0.25Vdd, 0.5Vdd and Vdd can be fed to it to check the output signals. If not in use, this pin must be tied to ground. Figure2.8 shows the sensing module and its PCB layout whose size is 25×20mm. This small module can be connected to specific receivers like DDS or any compatible devices.


Figure 2.8 The sensing module and its PCB Layout

2.2.3 Output response

Figure 2.9 illustrates the coordinates of acceleration in KXM52-1050, and Figure 2.10 and 2.11 show the response of this unit with respect to different directions of accelerations and with respect to gravitational direction in different positions, respectively. If there is no acceleration applied along an axis, the output voltage Voff equals half Vdd. If acceleration exists towards a positive direction, the output voltage increases (Vout> Voff) and vice versa.

The sensing range of KXM52-1050 of each axis is ±2g and the output varies with acceleration linearly at the rate of 660mV/g. Equation (3) shows the relation:


For example, when KXM52-1050 is put horizontally and still, there are no gravitational effects along X and Y axes, and VoutX=VoutY=Vdd. Only 1g vertical acceleration along the negative direction of the Z-axis is detected whose effect can be regarded as the accelerometer moving in the positive direction of Z-axis. In this condition, the output voltage at pin 14 is VoutY>Voff+660mV(at 3.3V Vdd).      


Figure 2.9 The coordinates of acceleration of KXM52-1050

Figure 2.10 Output response due to directions of accelerations

Figure 2.11 Output response due to gravitational directions in different positions

Tilt/inclination sensing is a common application for low-g accelerometers. Figure 2.12 indicates the tilt assignments (pitch and roll), where f, r, and q stand for the tilt angle with respect to X, Y and Z axes relative to the ground. Equation (4) identifies the relation between the tilt angles and accelerations of each axis.




Figure 2.12 Pitch and roll assignments relative to ground

A simple test is conducted when KXM52-1050 rotates about the X axis from 0o to 180o, and the data is shown in Table 1 and Figure 2.13. In this figure, the bold blue curve is obtained by actual test as compared with the narrow black curve which approximates a sinusoidal form generated according to the specification. Referring to Equation (3), 1g acceleration yields an output of about 2.31V, and 0 g yields about 1.65V. When f, r and q equal 0o, the outputs of the X and Y axes have the maximum sensitivity, while the Z axis has the minimum sensitivity. Occurrence of slight signal deviation is probably due to the offset effect caused by temperature variation and instrumentation variances. Figure 2.14 shows a linear response to gravity (acceleration) from 0g to 1g.

Table 1. Output voltages of KXM52-1050 w.r.t specific tilt angles (X axis)  




































































































Figure 2.13 Tilt angle vs. output voltage of X-axis from 0˚ to 180˚

Figure 2.14 Acceleration (0~1g) vs. output voltage 

2.3 Human body acceleration

2.3.1 Frequency and amplitude of the body motion

Frequency and amplitude of the body motion are two major components when analyzing acceleration data. In general, even for exercises such as running or jumping, frequencies and amplitudes of accelerations caused by body movements are relatively low. This implies that in data registration, it is not necessary to use high sampling rate. When walking at normal velocity, the frequencies are higher in vertical direction than that in both medio-lateral and antero-posterior directions. Meanwhile, Cappozzo’s research showed that the acceleration at head (cranial part) ranges from 0.8~5Hz whereas higher frequencies up to 60Hz may occur at the foot (caudal part) when the heel striking [Cappozzo, 1982]. Bhattacharya et al. [1980] found the majority of frequency components during running to fall between 1~18Hz in the vertical direction at the ankle. Although frequencies up to 60Hz have been reported in Cappozzo’s studies, these might be caused by external vibration as such high frequencies seem irrational in normal body movement.

The amplitude characteristics are similar to the frequency spectrum described above. The amplitudes are higher in vertical direction than in both medio-lateral and antero-posterior directions when walking and locomotion. The amplitudes increase from the head toward the caudal part. During walking, -0.3~0.8g in the vertical direction, -0.2~2g at the head and -0.3~0.4g at the lower back have been observed with a stereophotogrammetry in Cappozzo’s studies. During running or trampoline jumping, peak amplitude up to 12g at the ankle measured by skin-mounted accelerometers has been reported [Bhattacharya et al., 1980].

In practical acceleration registration, the frequency outputs from motion sensors are limited for desirable sensitivity and signal integrity. High frequency bandwidth yields a high sensitivity in response to body movements. Consequently, an appropriate frequency bandwidth of sensor outputs should be carefully selected to reduce interfering noises while preserving original acceleration patterns. Bouten et al. [1997] used a low-pass filter (20Hz-9dB/octave), and a high-pass filter (0.11Hz-0.6dB/octave) in order to eliminate DC response of piezo-resistive accelerometers. In general, the cut-off frequencies of the low-pass filters are below 50Hz.

Bouten et al. [1997] also suggested that body-fixed accelerometers must be able to register accelerations within the amplitude range of -12 to 12g and with frequency up to 20Hz in order to assess daily physical. High-g accelerometers were suggested since they have taken the activities such as trampoline jumping and running into considerations. However, for ambulatory monitoring at home environment, it is not necessary to use high-g accelerometers because these types of exercises are rare and the intensity is also quite low in such situations. Bijan Najafi et al. used two low-g accelerometers (ADXL202, ±2g) in the kinematic sensor for ambulatory monitoring.

 2.3.2 Sensor placement

The sensor placement on the human body is an important issue. Two major concerns are where and how to fix the sensors. The sensor should be firmly fixed to the human body to avoid the sensor from loosening, sliding and jolting which will cause artifacts in the sensor outputs. Therefore, a sensor attached to clothing is not feasible. The ideal mehtod is to fix the sensor directly on the skin of the human body because the dislocation of the sensor would be minimal, and the vibration or resonance induced by body movements would not significantly affect the sensor outputs. However, this attachment causes discomfort and inconvenience in use. A preferable way proposed in several studies is a wearable belt which carries sensors. This attachment is convenient to the users and will not significantly influence the physical activity in their activities of daily living.

Accelerations generated from all skeletal parts of a body such as the head, trunk and extremities differ from each other when most body movements occur. Meanwhile, different accelerations can be distinguished from several local measuring points within the same part of a body. For example, acceleration patterns from the sternum, waist or back are not always the same. To acquire detailed information of body movement and energy expenditure, it is necessary to place numerous motion sensors on each part of a body. However, such a method can not be applied in ambulatory uses due to the complex inter-instrumentation. In addition, not all the acceleration data are related to the analysis on physical activity or energy expenditure. Therefore, recent studies have been focusing on how to measure the data only with a few, or even a single sensor for specific purposes. Most studies have concluded that the sensor should be attached to the trunk since this part of the body represents most of the body mass. Cavagna et al. [1964] placed the accelerometer at the lumbosacral level on the back as close as possible to the center of gravity of the body, when studying external work in walking and running. As shown in Figure 2.15, Gerwin, Meijer, Carlijn and Bouten et al, used a belt with an accelerometer worn on the waist of the subject, located at the lower back (second lumbar vertebra) [Meijer et al, 1991; Bouten et al, 1997]. Najafi et al. [2002] used a piezoelectric gyroscope attached with a belt in front of the sternum to determine postural transitions, as well as to evaluate fall risk in the elderly. In their research for the ambulatory monitoring system, one kinematic sensor package was directly attached to the chest under the assumption that it will accurately detect postural transitions between standing, sitting, and lying, as well as locomotion activity when standing (Figure 2.16) [K.Aminian et al, 2003].

Figure 2.15 Sensor attachment with a belt on the waist

Figure 2.16 Sensor attached to the chest

2.4 Preliminary experiments on posture transitions

Preliminary experiments were performed on several normal adult subjects. The cut-off frequency of KXM52-1050 is 50Hz, and the following data acquisition was selected at the rate of 100Hz to preserve the signal integrity of A/D conversion. The accelerometer was attached firmly to the leather belt of the pants so that the attachment would not loosen or dislocate during the experiments.

2.4.1 Sit-to-stand and stand-to-sit posture transitions

Figure 2.17 illustrates the process of sit-to-stand posture transition of a typical tester. The reverse order in Figure 2.17 can also represent a stand-to-sit transition. As shown in Figure 2.18, when sit-to-stand or stand-to-sit posture transitions occur, there are few variations of accelerations along medio-lateral direction. Acceleration along this direction does not provide significant information for posture transitions, and therefore is not considered in this posture.

Figure 2.19 shows the tilt angle variation of the trunk with respect to vertical when sit-to-stand transition occurs. Initially, trunk tilt of about 25 to 30 degrees anteriorly is observed before standing up due to the autonomic balance for center of gravity of the body. In this phase the time duration is about 0.75 to 0.8 seconds. As the subject stands up from the seat, the trunk reverses gently to the original vertical posture within time duration of about 1.0 to 1.2 second.

Figure 2.17 Illustration of trunk tilt angle variation during sit-to-stand transition

Figure 2.18 Response in medio-lateral direction when sit-to-stand and stand-to-sit posture transitions.

Figure 2.19 Trunk tilt angle variation during sit-to-stand transition

To determine the acceleration pattern of posture transition, two channels of the KXM52-1050 are used to measure the accelerations along the vertical and antero-posterior directions (Figure 20). In the vertical component acceleration of about 0.14g upwards is detected when the subject stands up from the seat, and followed by a continuous deceleration up to -0.2g downward. The motion ends with a more gentle acceleration from -0.2g to 0g.

The antero-posterior component results from the trunk tilt forward and the following motion which generates a part of horizontal acceleration ahead. Acceleration up to about 0.4g is detected in this process. Note that the antero-posterior component begins and ends in about the same time as that along the vertical direction. Besides, signal outputs before and after posture transitions may not always be the same. This is because the initial and final orientation of the sensor may not be consistent due to the non-habitual posture of the subject or a slight dislocation of the sensor attachment exists.

Figure 2.20 Acceleration patterns of sit-to-stand posture transition

Similar characteristics of the acceleration pattern can be observed in stand-to-sit transitions. Figure 2.21 shows the trunk tilt. In the vertical component in Figure 2.22, an increasing acceleration downward begins with body motion when sitting down. In this phase 0.1g to 0.2g downward are measured. While the buttocks approach the seat, deceleration begins to slow down the motion. At the instant when the subject sits on the seat, a high, upward acceleration bouncing 0.2g up to 0.8g is detected due to the reacting force exerted by the seat. Note that the vertical bouncing acceleration upward may be different according to various materials of the cushion on the seat. A stiff cushion may results in higher values in contrast to lower values acquired with a soft cushion. Figure 2.23 reveals a high bouncing acceleration over 1g (up to 1.22g) when the subject sits on flat, stiff chair during stand-to-sit transition.

Figure 2.21 Trunk tilt angle of stand-to-sit posture transition

Figure 2.22 Acceleration patterns of a normal stand-to-sit transition

Figure 2.23 Acceleration patterns of a fast stand-to-sit transition

2.4.2 Walking and lie-to-sit transition

Figure 2.24 shows the acceleration patterns when walking. In this figure walking begins at T=0.5s. For the acceleration component in the antero-posterior direction, represented by the narrow red curve, each stride results in an increasing acceleration forward during the time interval of T=0.5s to T=1.1s. The maximum occurs at the end of this time interval. After that instant, the acceleration decreases as the foot is lowering down. As the foot steps on the ground whose peak value of vertical component can be detected, the time interval is approximately 0.1s to 0.15s. The maximal deceleration (negative peak) occurs after each step in a short time interval of about 0.1s.

For the vertical component represented by the bold blue curve, each step results in corresponding positive peak values about 0.3g to 0.45g. However, negative peak values are not obvious. Pairs of the two positive peak values whose vertical component is always behind the antero-posterior component in a narrow time intervals can be observed as indicated by the red arrows in Figure 2.24. When that foot steps on the ground in turn, the cycle described above re-appears. Acceleration patterns of faster walking pace registered in Figure 2.25 also shows the identical and definite characteristics. The only significant distinction reveals that faster walking pace results in higher positive peaks in vertical component. Note that in the negative regions of Figure 2.24 and Figure 2.25 the tendency of two pairs of negative peak values is not apparent compared with that in positive region as the two arrows indicate. Moreover, positive peaks values are higher than negative ones due to the effect of reactive force exerted by the ground.

Figure 2.24 Acceleration patterns of walking


Figure 2.25 Acceleration patterns of walking

Figure 2.26 illustrates the shift of trunk orientation which coincides with the sensor orientation. Figure 2.27 and Figure 2.28 show the gravity patterns of sit-to-lie or lie-to-sit posture transitions, these movements can be regarded as rotations of trunk with respect to the lumbar vertebra. Therefore, in sit-to-lie posture transition the vertical component of gravity decreases form 1g to 0g. Antero-posterior component (narrow line) of gravity varies from 0g to -1g. Patterns of lie-to-sit posture transition can be observed in reverse order.


Figure 2.26 Shift of sensor orientation when sit-to-lie posture transition

Figure 2.27 Gravity patterns of sit-to-lie transition

Figure 2.28 Gravity patterns of lie-to-sit transition

This section describes the preliminary tests on the acquisition of acceleration patterns of posture transition and walking. These postures and posture transitions can be clearly identified with several significant properties distinguished from one another. These results are of great help for the development of body movement algorithms which aimed to determine these postures or posture transitions.


Aminian, K., Najafi, B., Bula, C., Leyvraz PF., Robert, Ph., 2002. “Spatial-temporal parameters of gait measured by an ambulatory system using miniature gyroscope.” Journal of Biomechanics, vol. 35, pp. 689-699. 

Bhattacharya, A., McCutcheon E. P., Shvartz, E., Greenleaf, J. E., 1980. “Body acceleration distribution and O2 uptake in humans during running and jumping,” J. Appl. Physiol., vol. 49, pp. 881-887

Bouten, C. V. C., Koekkoek, K. T. M., Verduin, M., Kodde, R., Janssen, J. D., 1997. “A Triaxial Accelerometer and Portable Data Processing Unit for the Assessment of Daily Physical Activity,” IEEE Trans. Biomed. Eng., vol. 44, pp.136-147.

Cappozzo, A., 1982. “Low frequency self-generated vibration during ambulation in normal men,” J. Biomed., vol. 15, pp. 599-609.

Kemp, B., Janssen, A., Kamp. B. V. D., 1998. “Body position can be monitored in 3D using miniature accelerometers and earth-magnetic field sensors,” Electroencephalography and clinical Neurophysiology, vol. 109, pp. 484-488.

Meijier, G. A. L., Westerterp, K. R., Verhoeven, F. M., Koper, H. B. M., Hoor, F. T., 1991. “Methods to assess physical activity with special reference to motion sensors and accelerometers,” IEEE Biomed. Eng., vol. 38, pp. 221-229,

Moris, J. R., 1973. “Accelerometry-a technique for the measurement of human body movements,” J. Biomech., vol. 6, pp. 729-736.

Najafi, B., Aminian, K., 2002. “Measurement of Stand-Sit and Sit-Stand Transitions Using a Miniature Gyroscope and Its Application in Fall Risk Evaluation in the Elderly,” IEEE Trans. Biomed. Eng., vol. 49, pp. 843-851.

Schulman, J. L., Reisman, J. M., 1959. “An objective measure of hyperactivity,” Amer. J. Mem. Defic., vol. 64, pp. 455-456.

Zhu, H. S., Wertsch J. J., Harris, G. F., Loftsgaarden, J. D., Price, M. B., 1991. “Foot pressure distribution during walking and shuffling,” Archives of Physical Medicine and Rehabilitation, vol. 72, pp. 390-397.