Article
A Wearable System for the Evaluation of the
Human-Horse Interaction: A Preliminary Study
Andrea Guidi
1
, Antonio Lanata
1,
*
,†
, Paolo Baragli
2
, Gaetano Valenza
1
and Enzo Pasquale Scilingo
1
1
Research Center “E.Piaggio”, School of Engineering, University of Pisa, Largo Lucio Lazzarino 1,
Pisa 56122, Italy; andrea.guidi@for.unipi.it (A.G.); g.valenza@iet.unipi.it (G.V.);
e.scilingo@centropiaggio.unipi.it (E.P.S.)
2
Department of Veterinary Sciences, Laboratory of Equine Behavior and Physiology, University of Pisa,
Viale delle Piagge 2, Pisa 56124, Italy; paolo.baragli@unipi.it
* Correspondence: antonio.lanata@unipi.it; Tel.: +39-050-2217604
† Current address: Department of Information Engineering, School of Engineering, University of Pisa,
Via Caruso 16, Pisa 56122, Italy.
Academic Editor: Mostafa Bassiouni
Received: 26 June 2016; Accepted: 18 September 2016; Published: 26 September 2016
Abstract:
This study reports on a preliminary estimation of the human-horse interaction through
the analysis of the heart rate variability (HRV) in both human and animal by using the dynamic
time warping (DTW) algorithm. Here, we present a wearable system for HRV monitoring in horses.
Specifically, we first present a validation of a wearable electrocardiographic (ECG) monitoring
system for horses in terms of comfort and robustness, then we introduce a preliminary objective
estimation of the human-horse interaction. The performance of the proposed wearable system
for horses was compared with a standard system in terms of movement artifact (MA) percentage.
Seven healthy horses were monitored without any movement constraints. As a result, the lower
amount of MA% of the wearable system suggests that it could be profitably used for reliable
measurement of physiological parameters related to the autonomic nervous system (ANS) activity
in horses, such as the HRV. Human-horse interaction estimation was achieved through the analysis
of their HRV time series. Specifically, DTW was applied to estimate dynamic coupling between
human and horse in a group of fourteen human subjects and one horse. Moreover, a support vector
machine (SVM) classifier was able to recognize the three classes of interaction with an accuracy
greater than 78%. Preliminary significant results showed the discrimination of three distinct real
human-animal interaction levels. These results open the measurement and characterization of the
already empirically-proven relationship between human and horse.
Keywords:
wearable systems; e-textile; human interaction; biomedical signal processing;
non-stationary signal
1. Introduction
In the last few decades, the interest in decoding the human-horse relationship and interaction
(HHRI) has increased dramatically. This was guided by the strong empirical evidence of the positive
outcomes in equine assisted therapy (EAT) and horseback riding in therapeutic programs [
1
], as well
as the positive impact of animal companionship on human quality of life [
2
], where the equine
is an important element of these therapeutic practices, with its feelings and behavior. For this
purpose, the investigation of the modality in which both human and horse can communicate might
be crucial. Measuring and evaluating the impact of the interaction experience might be relevant [
3
].
Some studies investigated the equine perception of humans in terms of positive, negative or neutral
Electronics 2016, 5, 63; doi:10.3390/electronics5040063 www.mdpi.com/journal/electronics
Electronics 2016, 5, 63 2 of 17
valence [
4
]. A study on how human psychological and physiological state can be perceived by horses
was performed in [
5
] via the study of the heart rate. A more relaxed equine behavior was observed
when humans showed positive attitudes toward them [
6
,
7
], while an equine increased heart rate was
observed when humans were engaged in negative thinking [
8
]. A nervous mood can be transmitted
from humans to horses under handling and riding conditions [
9
].
Voice [10–12]
, posture [
13
,
14
],
facial expression
[15,16]
, autonomic
signals [17–19]
, hormones [
20
–
23
] and pheromones [
24
] might be
used to fruitfully describe and characterize the emotional content [
25
]. Non-verbal communication
between human and horse was also investigated in [
26
,
27
]. Heart rate and behavior resulted in being
sensitive and reliable indicators of fear or anxiety in horses [
28
,
29
]. Horses that are in discomfort
were observed to be more aggressive toward humans [
30
] or to be characterized by an increased heart
rate, motor activity and vocalizations [
31
]. The effect of the gender of the person interacting with
the horse was discussed in [
32
]. Although a recent review described a parallel behavior between
the human multi-sensorial perception and the demonstrated equine cross-modal
recognition [33,34]
,
an interdisciplinary approach is mandatory to reach a deeper knowledge of human-horse interaction.
In fact, these achieved experimental findings pointed out complex and multidimensional aspects of
the interaction, which involve medical, bioengineering, physics and veterinary science [
2
,
35
,
36
]. In [
9
],
the heart rate of both human and horse were monitored simultaneously under different experimental
handling or riding conditions. In this study, Keeling et al. asserted that the analysis of heart rate is
an important tool to investigate horse-human interactions. Again, hormones, heart rate and some
standard heart rate variability-related indices were investigated in [
20
] during both training and
performance. Different feature trends were observed between human and horse when they were
obtained from ECG (electrocardiographic) recordings related to public or private sessions. A body
sensors network technology was used to real-time monitor the horse-rider dyad in [
37
]. The aim
of such a monitoring was evaluating the human-horse interaction. Specifically, based on a study
concerning the equine emotional response during physical activity [
38
], the evaluation was based
on the measurements of heart rate and physical activities via a mathematical model. Such a model
was proposed to decompose the equine heart rate into two different components. The first one
was concerning the physical component, while the second one contained information about equine
emotional state [
38
]. Finally, in [
39
], the link between horse and human was also investigated by
studying their electroencephalograph signals (EEG), revealing a higher synchronicity in EEG waves at
increasing interactions. A correlation analysis between human and equine hormone concentrations
was performed in [20,21].
Therefore, the human-horse relationship appears to be a complex interaction affected by several
psychological factors [
40
]. The perception of humans by horses seems to be based on experience
and repeated interactions, with horses that form a memory of humans that impacts their reactions
in subsequent interactions [
5
]. Hence, previous negative experiences with human contact could lead
horses toward a negative emotional reaction [
7
] or, vice versa, previous positive experience could lead
them toward positive feelings with humans [41].
In our hypothesis, human and horse are considered as complex systems that interact through
a coupling process. In this frame, we hypothesize that coupling can be modulated by the type and
time duration of the contact itself. Specifically, we analyzed the level of coupling by studying their
heart rate variability (HRV) time series [
42
] through dynamic time warping (DTW). As is well known
from the literature, HRV can be considered as a non-linear time series, in which complex oscillations
are present [
43
]. Therefore, we aimed at measuring this biological coupling over time [
44
]. DTW is,
generally, used for studying time series dynamics of non-stationary systems. It calculates the best
possible warp alignment between two time series, by selecting the one with the minimum distortion.
Specifically, DTW is also defined as a measure of similarity between two time series, and it is calculated
as the minimum mapping distance between them. DTW was widely used in many contexts, including
data mining [45], speech processing [46] and medicine [47].
Electronics 2016, 5, 63 3 of 17
In this study, DTW was adopted to evaluate how heart activities evolve in a similar or dissimilar
way. For example, if during an experiment with increasing exciting levels we detect an increasing DTW
between human and equine HRVs, it indicates that the distance of the two HRV dynamics is increasing,
and therefore, the heart activities are following different patterns. In order to perform a continuous,
comfortable and non-invasive monitoring of the interaction in a natural environment, we developed,
and here present, a wearable monitoring system for horses. The amount of advantages that wearable
systems have brought to physiological signal monitoring for humans is well known, occupying an even
larger space in the research. Moreover, the continuous technological development and the increasing
demand of smart systems push wearables as the most used and suitable systems for ubiquitous and
pervasive investigations. In addition, their flexibility allows researchers and clinicians to face the large
variability of biomedical signals and tasks in monitoring subjects during their daily activities [
48
–
52
].
However, the biggest limitation in using wearable systems with humans and animals is due to motion
artifacts (MA) [
53
–
58
], which are the major source of noise in biomedical signal acquisition, inducing
the loss and alteration of informative content. For instance, the electrocardiographic (ECG) signal
acquired in a naturalistic environment without movement constraints can be severely affected by
important artifacts, and a great amount of data might be lost in contrast to the signal quality easily
achievable in controlled environments and protocols [
59
]. As a matter of fact, cardiac stress tests
or simply respiration can generate a big amount of MAs that can alter the signal [
59
]. Moreover,
it is important to highlight that restraining horse is usually discouraged since it is unnatural and
stressful and induces an increase of the sympathetic contribution of the heart control [
60
] that leads
to misleading ECG interpretations [
61
]. In this work, we present a textile-based system for the ECG
monitoring in horses, where the electrodes are completely made of fabric
(electro-textile or e-textile)
.
Normally, textile materials are insulators, but for this application, conductive yarns are integrated
into the fabric during the manufacturing process [
62
]. In the human biomedical field, e-textiles are
considered as higher value-added textiles and are prominently developed for being used in smart
clothing. Smart clothing refers to a new garment that is able to acquire and process information, as
well as actuate responses [
63
]. The potentialities deriving from these kinds of textile sensors enable
the application of wearable systems in a great variety of experimental settings. As a matter of fact,
many human studies showed reliable recordings of biomedical signals [
64
–
68
]: for example, ECG
in [
69
–
76
], respiration in [
69
,
70
,
73
,
74
,
77
–
79
], electrodermal response (EDR) in [
51
,
80
] and, finally,
PhotoPlethysmoGraphy (PPG)and blood pressure in [
81
,
82
]. In this kind of application, physiological
signals are monitored and recorded in order to evaluate or follow the health status of the person who
is wearing the wearable device [48,83].
Similarly in the veterinary research field, some authors [
84
–
87
] proposed to use Holter devices
in equine applications and subsequently systems with radiotelemetry. Here, we propose the use
of wearable systems in both animals and humans simultaneously, in order to acquire their cardiac
activity in a reliable and artifact-free way, as was generally demonstrated in [
72
,
88
–
92
]. The aim is
to infer autonomic nervous system responses enabling the detection of uncontrolled responses of
animals when elicited with emotional stimuli coming from humans and vice versa. The manuscript
is organized as follows: Section 2 deals with the materials’ and methods’ description; specifically,
it describes the wearable systems, the experimental protocol of the interaction and the signal processing
chain. Section
3 reports the achieved results, and Section 4 is focused on the discussion and conclusion
of the study highlighting the future perspectives of this pioneering work.
2. Materials and Methods
In this study, two wearable monitoring systems, i.e., one for humans and the other for horses,
were employed. Each system was comprised of a smart garment and portable electronics. The human
system (Figure 1) was designed as a sensorized t-shirt (Smartex, Pisa, IT, Italy), and it is exhaustively
described in [
48
,
79
,
93
–
95
]. Differently, the equine system (Figure 2) was comprised of an elastic smart
belt (Smartex, Pisa, IT, Italy) [
89
,
91
] fastened around the chest behind the shoulder area. In both
Electronics 2016, 5, 63 4 of 17
systems, two textiles electrodes (Smartex, Pisa, IT, Italy) and a strain gauge sensor (Smartex, Pisa,
IT, Italy) were integrated to acquire the ECG (with a sampling frequency of 250 Hz) and respiration
activity (with a sampling frequency of 25 Hz). The strain gauge is carried out by textile sensors that
monitor the cross-sectional variations of the rib cage. The respiration sensor along with electrodes
are integrated in an elastic band through a one-step process in the fabric by means of a circular
knitting machine [
96
]. They are developed by Smartex s.r.l.; many details can be found in [
92
,
96
].
Moreover, equine ECG was recorded by placing the electrodes according to the modified base-apex
configuration [
97
]. It is worthwhile to note that the use of a dry textile-based electrode provides
several advantages. Firstly, the system is easy to use through an automatic placement of the sensors
and allows high comfort. Secondly, electrodes are made of a special multilayer structure of textile
material that increases the amount of sweat and reduces the rate of evaporation reaching very rapidly
an electrochemical equilibrium between skin and electrode. This means that the signal quality [
72
] is
remarkably improved and kept as constant as possible. These materials are knitted together and are
fully integrated in the garment without any mechanical and physical discontinuity, creating areas with
different functionalities (see Figure 1). For each system, the two ECG e-textile electrodes and the strain
gauge sensor are finally connected to portable electronics through a simple plug that can be easily
unplugged when necessary.
Figure 1.
Wearable monitoring system for humans [
48
]. As is possible to note, from the box on the
right, the e-textile electrodes for ECG (electrocardiographic) acquisition are knitted and completely
integrated into the garment.
Portable
Electronics
Ag/AgCl
Electrodes
H-textile
Electrodes
Elastic belt
Strain gauge
sensor
Figure 2.
Systems placement on the horse. Elastic smart belt and Ag/AgCl electrodes placement.
A scheme is presented on the left side and a real picture is on the right side.
Electronics 2016, 5, 63 5 of 17
Moreover, an inertial platform (triaxial accelerometer) (Smartex, Pisa, IT, Italy) integrated into
the portable electronics that was positioned on the back of the horse allowed the monitoring of the
horse’s physical activity at a sampling frequency of 25 Hz. Finally, the wearable system was wirelessly
connected to a smartphone where a dedicated app enabled checking the status of the sensors and
remotely controlling the storing process of the physiological information in a secure digital (SD) card.
2.1. Equine Textile-Based ECG: Test and Validation
This section is focused on testing and validating the reliability of the equine ECG traces coming
from textile electrodes. For this study, seven healthy standard bred mares in anestrus (mean age
8.4 ± 1.3
years) were enrolled. Equine subjects were socially housed in a paddock (75
×
75 m) and
were provided ad libitum access to both hay and water. Horses were used as receivers in the embryo
transfer program of the Department of Veterinary Sciences (University of Pisa, Italy) where this study
was performed. Mares were in healthy condition and not pregnant at the time of this protocol.
The signal quality test of e-textile electrodes was performed by means of a comparative study in
terms of the motion artifact (MA) between e-textiles and conventional Ag/AgCl electrodes. Each horse
was simultaneously monitored by the wearable system, i.e., equine elastic smart belt and Ag/AgCl
electrodes (see Figure 2). Two identical electronics were employed for both couples of electrodes
Ag/AgCl and the textile ones. All data were acquired in a stall (4
×
4 m), where horses were left
free to spontaneously move for 60 min. At the end of the recording session, all of the acquired ECGs
were visually examined by one expert, and all of the ECG segments, corrupted by MAs, were marked.
The goal was the estimation of the percentage of corrupted signal. Such a percentage was calculated
as the sum of the time intervals in which an MA was observed over the whole time length of the
recording.
A nonparametric Wilcoxon signed-rank test for paired samples was used to compare the
percentage of corrupted signals between the signals acquired by the two kinds of electrodes,
i.e., e-textile and Ag/AgCl. Specifically, the test was designed to compare the performance of the
two kinds of electrodes while simultaneously recording the heart activity of each horse. A significant
result of such a test would indicate a coherent different percentage of corrupted signal between the
traces acquired by means of the two kinds of electrodes.
2.2. Protocol of Interaction
Here, the design of the interaction protocol is reported. Fourteen subjects (25
±
3 years old,
4 males) were enrolled. The participants did not show any past or current experience of mental or
personality disorder. In addition, one standard bred mare out of the seven previously described
was enrolled (age: 8 years, weight: 560 kg, height: 160 cm). Informed consent was signed by the
participants according to this specific protocol approved by the Ethical Committee of the University
of Pisa.
During the whole experimental protocol, the autonomic nervous system (ANS) response of both
human and horse were monitored. Specifically, two different systems were used to acquire the ECG
signal. The fabric-based monitoring system [
48
,
79
,
93
–
95
] was used to record the human ECG, while the
elastic smart belt [89,91] was used to record the equine one.
The experimental protocol consisted of three phases, each one lasting 4 min. During the first phase,
P
1
, the human and the mare were in different stalls (4
×
4 m). In this phase, considered as the resting
phase, the subject sat on a chair, while the horse was free to move. Successively, the horse was moved
from her stall to the human’s. In this phase,
P
2
, the subject was asked to keep himself/herself still on
the chair, while the horse was free to move and explore the environment. This phase implies a visual
and olfactory interaction. Finally, in the third phase,
P
3
, all of the participants were asked to stand up
and groom the horse. It is worthwhile to note that in order to keep the visual conditions of the horse as
constant as possible, all of the subjects were asked to wear an azure plain t-shirt.