The hand is an amazing feat of human evolution enabling incredible physical dexterity and the ability to manipulate and develop tools through a series of finger motions, i.e., flexion, extension, abduction, adduction, and circumduction. Finger flexion is joint motion towards the palm relative to the standard anatomical position with extension being in the opposite direction as shown in Figure 1
. Full extension is in line with the back of the hand and is defined as zero degree flexion. An injury can occur when joints are hyperextended too much [1
]; in this paper, only hyperextension within normal limits is of interest. Finger abduction (ABD) and adduction (ADD) are the movements away from or towards the hand’s midline (dotted line), respectively. Circumduction is a circular motion combining flexion, extension, abduction and adduction [2
Understanding and measuring hand kinematics is important in research areas such as biomechanics and medicine. It is a challenging subject due to the articulated nature of the hand and the associated multiple degrees of freedom (DOFs). Standard clinical goniometers are typically manual devices used to assess the range of joint movement in fingers [3
]. This method is only effective for static measurements and usually requires a long time to fully assess the whole hand. The accuracy (±5
) is largely determined by the skill of the clinician or therapist; therefore, an accurate system that is able to monitor dynamic hand movements could be of significant benefit in clinical practice.
Various sensing techniques have been proposed to track finger motion, including resistive carbon/ink sensors [5
], microfluidic strain sensors [7
], magnetic induction coils [9
], hetero-core spliced fibre sensors [10
], fibre Bragg grating sensors [11
] and multiple inertial measurement units (IMUs) [13
]. Resistive film sensors are particularly favoured in glove-instrumented devices due to their flexibility, light weight, and low cost; however, they often suffer from instability, long transient response, and a high dependency on the radius of curvature [15
], resulting in low accuracy. Previous work has shown commercial resistive bend sensors (Flexpoint Sensor Systems Inc., Draper, UT, USA [16
]) suffer from severe hysteresis error and signal drift compared to our optical-based sensor [17
]. In contrast, magnetic techniques are capable of monitoring precise hand movements, but the accuracy is significantly affected by interference from the Earth’s geomagnetic field or nearby ferromagnetic objects [19
]. Optical fibre techniques measure finger flexion by detecting the attenuation of transmitted light as a function of fibre curvature; a polishing process [20
] or the creation of imperfections [21
] is required to increase sensitivity.
Using one or more of these sensing techniques, several glove-based devices have been developed to capture hand movements, e.g., SIGMA glove [22
], SmartGlove [23
], WU Glove [24
], NeuroAssess Glove [25
], Shadow Monitor [26
], HITEG-Glove [27
], Pinch Glove [28
], Power Glove [29
], CyberGlove [30
], HumanGlove [31
], and 5DT Data Glove [32
]. These glove-based devices in most cases only detect finger flexion and/or extension using one-DOF sensors [10
]. The measurement accuracy is usually adversely affected when monitoring two-DOF finger joints, i.e., the metacarpophalangeal (MCP) joints. The main reason is that finger ABD/ADD and flexion/extension do not occur in isolation [35
], leading to unwanted deformations of the flexion sensors. Few glove systems have been able to measure two-DOF finger movements. One method is to use multiple one-DOF flexible sensors. The flexion/extension sensors are placed on the finger joints, and the ABD/ADD sensors are attached on the dorsal surface of the proximal phalanxes of adjacent fingers in an arched configuration [22
]. In this approach, reliable ABD/ADD readings can only be obtained when the adjacent fingers have similar flexion; otherwise, twisting of the ABD/ADD sensor can occur, leading to inaccurate readings. Gestures, such as crossed fingers, are particularly prone to this problem. An alternative method is the use of IMUs that can provide 3D information of hand motion including finger ABD/ADD. The major drawback is the accumulated error since IMUs estimate the orientation trajectories by time integration of the inertial signal. Additionally, complicated models and data fusion processes are also required [14
], causing large computation times for multiple joint articulations, thus limiting the system’s real-time capability.
Both finger flexion/extension and ABD/ADD are important parameters for finger kinetics. The majority of studies to date only consider finger flexion and extension, but accurate information on finger ABD/ADD is essential for areas such as medical diagnostics and rehabilitation. Accurate and simultaneous monitoring of two-DOF motion in human hand joints is an area underexplored in current glove-based systems. The main purpose of this work is to demonstrate a method for tackling this problem. We propose a two-axis optical sensor with an operating method based on Malus’ Law, capable of tracking the finger flexion/extension and ABD/ADD simultaneously. Additionally, the sensor demonstrates an ability to capture the finger circumduction, which is important for exploring the complex relationships between finger flexion/extension and ABD/ADD. This optical sensor features a wide measuring span of up to 180 for each axis, which is sufficient for measuring the entire range of finger motion. In the following sections, the hand skeleton model, sensor principle and fabrication, quasi-static characteristics, as well as the recorded movements of the MCP joint of the left index finger will be described.
2. Hand Skeleton Model
The human hand is highly articulated, but constrained at the same time. Each hand has up to 27 DOFs, 21 of which are contributed by the five finger joints for local movements and the other six for global hand movements [35
]. Each digit, except the thumb, possesses four DOFs, one each for the proximal interphalangeal (PIP) and distal interphalangeal (DIP) joints, and the other two for the MCP joint. The thumb has a more complicated structure, possessing five DOFs, one for the interphalangeal (IP) joint, two for the MCP joint, and two more for the carpometacarpal (CMC) joint [35
Six types of synovial joints typically exist in the human body with differing structure and mobility [38
] (pp. 18–20): plane joints, hinge joints, pivot joints, ellipsoidal joints, saddle joints, and ball and socket joints. Human fingers only have hinge, ellipsoid and saddle joints. The uniaxial and stable hinge joint allows flexion/extension only, performing back and forth movements. The PIP, DIP, and IP joints are examples of the hinge type [38
] (pp. 170, 174). Ellipsoidal joints, also known as condyloid joints, consist of two oval articular surfaces; one is concave, and the other one is convex. This type of joint is able to move in two planes, allowing flexion, extension, abduction, adduction, and circumduction. MCP joints are examples of the ellipsoidal type [38
] (p. 169). Saddle joints are also biaxial and perform a series of movements, similar to ellipsoidal joints, utilizing two reciprocally concave–convex surfaces. The CMC joint of the thumb is a typical saddle joint [38
] (pp. 164–167).
In this paper, we focus on detecting the motion of the MCP joint. The range of motion varies between individuals [39
], but, in general, flexion is around 90
, and slightly less for the index finger [35
]. The active hyperextension of the index finger is approximately up to 30
. ABD/ADD generally have a greater range when the fingers are fully extended, being as much as 30
in each direction [38
] (p. 174).
5. Sensor Characteristics
The quasi-static performance of the two-axis optical sensor is evaluated using the automated measurement apparatus at room temperature. The motor performs rotations at 10 degrees per second and the data is acquired at 100 Hz. The physical position when the two wings of the sensor are aligned is defined as the origin.
To investigate the sensor’s angle-to-voltage relationship, the optical sensor was rotated, about the x-axis, from −100 to 100 and then back to −100 using increments of 5. The data set was composed of 500 samples at each angle setting. This process was repeated five times with an interval of three minutes between each sampling cycle. The same procedure was carried out for the rotation testing about the y-axis.
By averaging the data from CH
at each angle over five cycles, the voltage-to-angle relationship of both channels was obtained for both clockwise and anticlockwise rotations. The angular dependence of the output voltages is plotted in Figure 5
. According to Equation (2
are the output span and the offset voltage (the minimum output) of each channel CH
= 1, 3). The predicted values for CH
are calculated and also plotted in Figure 5
have a similar performance to that of CH
and have been omitted from Figure 5
As seen in Figure 5
, both channels provide readings consistent with the predicted values for bidirectional rotation. The phase difference between the waveforms is consistent with the difference in orientation of each channel.
The sensor’s output characteristics are listed in Table 1
. It demonstrates consistent performance between each channel with an average deviation from the theoretical voltage, equal to ±2.3%. The overall hysteresis error is less than 1.7% for rotation in both directions and the repeatability, as determined from the averaged relative standard deviations, is equal to 0.5%.
In the linear region, where
is in the range 20
, the sensitivity of each channel is 62.6 mV/
± 0.9 mV/
; this can be further increased by adjusting the amplification factor of the conditioning circuits. According to the signal-to-noise level, the resolution is 0.1
under laboratory conditions for the linear regions, comparing favorably with the 0.5
achieved in commercial resistive bend sensors [25
An optical sensor has been developed that is capable of capturing the complex motion of the human hand. The sensor uses a two-axis measurement technique, allowing the motion of two-DOF joints to be accurately recorded. This sensor has an accuracy of ±0.3, much higher than optical fibre and potentiometer based technologies. Additionally, a particular advantage of our sensor is its wide measurement range of 180 which is sufficient to track the entire range of motion of human fingers. Time-consuming calibration is also unnecessary for this sensor.
When attached to the MCP joint of the index finger, the sensor has successfully tracked finger motion in real time, including flexion/extension, ABD/ADD and circumduction. The optical sensor achieved angular measurements within 2.7 of the reference values obtained photogrammetrically. The promising results show the capability of the presented sensor for tracking two-DOF finger motion. This two-DOF optical sensor should also work for both PIP and DIP finger joints, since they have only one DOF and therefore limited mobility compared with the MCP joint. Furthermore, this sensor has the potential to be fabricated with other dimensions for monitoring other joints in the human body, e.g., ankle, wrist, and knee.
In this work, the optical sensor was attached to the hand with adhesive plasters instead of using gloves. This improves the measurement accuracy but may be unsuitable for long-term monitoring. Only the left index finger of one test subject was investigated here, so additional studies on multiple participants and multiple finger joints will be required to thoroughly validate the sensor’s reliability and reproducibility. Moreover, further reduction in size may be necessary to avoid collisions when multiple sensors are placed close to each other on the hand. These problems will be addressed in future work.