Unobtrusive Health Monitoring in Private Spaces: The Smart Vehicle

Unobtrusive in-vehicle health monitoring has the potential to use the driving time to perform regular medical check-ups. This work intends to provide a guide to currently proposed sensor systems for in-vehicle monitoring and to answer, in particular, the questions: (1) Which sensors are suitable for in-vehicle data collection? (2) Where should the sensors be placed? (3) Which biosignals or vital signs can be monitored in the vehicle? (4) Which purposes can be supported with the health data? We reviewed retrospective literature systematically and summarized the up-to-date research on leveraging sensor technology for unobtrusive in-vehicle health monitoring. PubMed, IEEE Xplore, and Scopus delivered 959 articles. We firstly screened titles and abstracts for relevance. Thereafter, we assessed the entire articles. Finally, 46 papers were included and analyzed. A guide is provided to the currently proposed sensor systems. Through this guide, potential sensor information can be derived from the biomedical data needed for respective purposes. The suggested locations for the corresponding sensors are also linked. Fifteen types of sensors were found. Driver-centered locations, such as steering wheel, car seat, and windscreen, are frequently used for mounting unobtrusive sensors, through which some typical biosignals like heart rate and respiration rate are measured. To date, most research focuses on sensor technology development, and most application-driven research aims at driving safety. Health-oriented research on the medical use of sensor-derived physiological parameters is still of interest.


Introduction
Unobtrusive and continuous health monitoring in private spaces uses sensor technology without introducing any inconveniences to the patient's normal life [1]. With respect to "private spaces", we refer to a living environment with limited public access, such as a home, apartment, or a privately owned car [2]. In sensor-enhanced private spaces, health-related information is collected continuously and critical changes or events are captured automatically. Furthermore, the collected information reflects the natural reality and promising services, including emergency detection, disease management, and health status feedback, and advice becomes achievable [3].
The Internet of Things (IoT) facilitates the ubiquitous sensing of all aspects of people's lives, including health, entertainment, activities, and communication [4]. Big data (variety, velocity, volume) is collected unobtrusively and artificial intelligence (AI) is applied to early detect diseases or predict health status [5,6]. As of today, people's private spaces are equipped with advanced technology, which is reshaping their lifestyles. For example, smart wearables track activities. AI-based personal assistants, such as Amazon Alexa or Google Home, have not only changed human-machine interaction but established these technologies as part of regular life [7], which also includes, for instance, autonomous driving.
Personal mobility is a distinctive trait of modernity. In many countries, people spend a considerable amount of time in cars: the average daily time is about 1 and 1.1 h in the United States [8] and in Germany [9], respectively. Furthermore, the private vehicle is one of the best-equipped environments in our daily life. This provides a great opportunity to convert the vehicle into a health monitoring facility [2] and to use the time people spent in their cars for health monitoring [10,11].
For many years, in-vehicle health monitoring has been the focus of research. This includes environmental, physiological, and behavioral monitoring: • Environmental parameters include temperature, air quality, humidity, weather and light conditions, and speed. They are captured already by default for in-car well-being and driver's assistance systems. • Physiological parameters typically include vital signs; in particular, heart rate (HR), respiration rate (RR), body surface temperature, and skin impedance. More advanced parameters can be measured with special sensing devices.

•
Behavioral parameters quantify physical activities during the drive to reflect the driver's attention level, tiredness, and well-being.
The use of camera sensors allows not only to directly observe the driver's activities [14] but also delivers vital signs. For instance, oxygen saturation is measured by attaching a light-emitting diode (LED) and photo-diode to the steering wheel [15] as well as from analyzing images captured by a camera attached to the windscreen or the control panel dashboard [16]. The RR can be measured when deploying a piezoelectric sensor or an accelerometer on the seat belt [11,15] but also from radar signal [17].
To date, several authors use physiological information to monitor the driver state, detect fatigue, or assess stress, and the data is fed into driving assistance systems [18][19][20]. These applications require robustness during the entire drive, which is hardly reached.
Medical applications, in contrast, profit from the regularity of every-day use and cope with dropouts during the ride. However, existing research does not provide a guide to currently proposed sensor systems for in-vehicle monitoring. In particular, the following questions need to be answered.
1. Which sensors are suitable for in-vehicle data collection? 2. Where should the sensors be placed? 3. Which biosignals or vital signs can be monitored in the vehicle? 4. Which purposes can be supported with the health data?

Methods
In this paper, we focus on unobtrusive continuous health monitoring in a smart vehicle, which we consider a private environment or a private space. We present terminology for sensors, locations, biosignals, and purposes and applied a comprehensive literature review to answer the stated questions.

Terminology of Unobtrusive In-Vehicle Health Monitoring
The primary goal of driving is to reach a predefined destination. The driver perceives information about the environment, such as weather, road conditions, and traffic signs, and accordingly controls the vehicle [21]. Driving consist of tasks on different levels [22]: • Strategic tasks (e.g., choice of route); • Navigational tasks (e.g., adherence to the chosen route); • Traffic-related tasks (e.g., interacting with other road users); • Adherence to rules (e.g., traffic signs and signals); • Tasks related to the road (e.g., chosen position within traffic); and • Speed control (e.g., maintenance of the speed according to road situation).
For these tasks, objects such as steering wheel, windscreen, mirrors, pedals, or the dashboard (speedometer) are used, and their context, location have to be considered [23,24]. Accordingly, we propose a flat terminology that covers unobtrusive sensors (electromagnetic, mechanic, optic), their locations (seat, chassis, instruments), and the biosignals or vital signs (body, heart, blood, lung, eye) that can be obtained by the sensors and the purposes ( Figure 1).

Literature Retrieval
Using our terminology of unobtrusive in-vehicle health monitoring (Figure 1), we developed the search string that reflects two aspects:

•
Biosignal consists of general terms, such as biosignal, biological signal, physiological signal, physiological parameter, vital signal, vital sign, vital parameter, and commonly seen specific biosignals terms, such as ECG, electrocardiograph, heart rate, heart rate variability, heartbeat, respiration rate, breathing rate, breathing, body movements; • Vehicle consists of terms regarding the vehicle, such as car, vehicle, automobile, automotive, drive, driving, driver.
We connect the terms within and across each aspect with logic OR and AND, respectively (Appendix A). We applied the search string to the title of articles from three databases: PubMed, IEEE Xplore, and Scopus on 16 December 2019. To reflect only up-to-date research, the publishing date spans the last decade (2009-2019). Furthermore, we restricted responses to the English language. After we combined all returned records, we removed duplicated papers. Then, we screened titles and abstracts according to Section 2.3 and excluded irrelevant records. Afterward, we studied the full papers and excluded work that was published already in included articles (in such cases, we found a large overlap in the content of papers of the same authors).

Review Criteria
We reviewed the retrieved full papers to identify research focusing on the application of sensor technology for unobtrusive health monitoring in the vehicle. As several persons performed the two-stage review, we defined the following criteria: • Inclusion -Unobtrusive sensors are part of the method; - The sensors are used to collect heath-relevant data, i.e., behavioral or physiological parameters.

-
The sensors are body-attached, wearable, or implanted; -Sensor data is not used for biosignal or health state monitoring; - Research is not on humans.
When analyzing the full texts, we extracted the type of sensor and its location, the targeted biosignals or medical parameters, and the main purpose of the research. Possible options for the purposes are:

•
Sensor development for measuring a certain health parameter; • Application of sensor data for health (i.e., disease management, diagnostics, prediction) or safety ( Figure 1).
With respect to the conditions of an experiment, possible options are: • On-road driving: the experiment was performed with naturalistic driving, where the subjects were required to drive a car on real roads; • Driving simulator: the subjects were required to simulate driving activities on a driving simulator; • Laboratory setting: a driving-like setting up or a separated (part of a) vehicle was equipped with sensors (e.g., seat, steering wheel), but no driving activity was simulated.
We further tracked the number of subjects that participated in the experiment.

Results
The initial search query on PubMed, IEEE Xplore, and Scopus resulted in 959 records after removing duplicates, of which 49 papers remained after assessing titles and abstracts ( Figure 2). When reviewing the full texts, we excluded three papers additionally, due to a high overlap in content to already included papers of the same authors (double publishing). Finally, 46 papers remained for text analysis (Table 1).
Based on the review criteria (i) sensor development vs. application and (ii) laboratory setting vs. diving simulator vs. on-road driving, we categorized all papers into six groups ( Figure 3).  The majority of research (n = 35) focuses on the development of sensor technology. Only a few works (n = 11) are toward problem-solving. Fourteen of the 35 papers for sensor development have conducted on-road driving tests, but most of them with only a few subjects (1)(2)(3)(4)(5). Exceptions are the work of Kuo et al. [25] and Lee et al. [26] with ten subjects in an on-road drive test. Nine out of 11 application-oriented research focus on safety, and only two papers on health-related issues.
According to our terminology ( Figure 1), we depict the state-of-the-art for in-vehicle health monitoring in a graph that links columns for location, sensor, biosignal, and purpose ( Figure 4). The interconnections are based on all 46 papers. In other words, we draw interconnections only on evidence in the literature. Also, we labeled the interconnections with the reference numbers of the corresponding papers. To enhance readability, we collected all interconnections and their supporting literature separately in Table 2.   Table 2).

Sensors
In total, 15 types of sensors were found in the included papers. Contact (dry) electrode, capacitive electrode, radar sensor, and video camera are the most popular sensing devices, which can be found in 11,9,8, and 7 papers, respectively. The papers involving the contact or capacitive electrodes were mostly published from 2009 to 2014 (n = 18), while the video-related work was mainly published from 2014 to 2018 (n = 8). Several authors collect ECG, EMG, or galvanic skin response (GSR) through electrode and corresponding amplifiers and filters [15,20,27,29,30,37]. When cameras are in use, the authors target PPG-derived HR or heart rate variability (HRV) [25,44,46,53,61]. Some work uses cameras naturally for activity or emotion monitoring [28,43]. The pulse oximeters are attached to the steering wheel and measure HR or RR [15,27,66]. BCG sensors are widely explored for in-vehicle scenarios. HR and RR are the typical health parameters that can be extracted from BCG [28,31,57]. In recent years, a radar sensor has become a device for recording physiological parameters, such as HR and RR [45,48,55,62,64]. Furthermore, two teams use vehicle-built-in sensors, such as GPS, to assess the driving behavior or the driver's mental health [49,58].

Locations
All included research assumed that the driver was the monitored person, i.e., driver-centered. Although there are plenty of options for human-vehicle interaction, only a limited number of locations are actually used for health monitoring. Car seats (n = 20) and steering wheels (n = 18) most frequently host sensing devices. The car seat (backrest and seating area) is a suitable location for capacitive electrodes [20], BCG sensors [31,57], magnetic induction sensors [33,51] and radar sensors [45,48].
Since the hands are the unique body part of the driver that directly contacts the vehicle, the steering wheel is equipped with contact electrodes, a pulse oximeter, and a thermometer [27,28,30,66]. The control panel, windscreen, and the windscreen-mounted rear-mirror mostly host a camera [25,53,56,61] or a gas sensor [50]. A modern vehicle is equipped already with many sensors for drive-train control, safety, and comfort [67]. The data from these built-in sensors are potential resources for behavior monitoring, and current research considers respective data interfaces (on-board diagnostics) [49,58].

Biosignals
HR (n = 24) and RR (n = 15) are the most frequently measured health parameters. HR data is often measured indirectly based on ECG, BCG, remote PPG (rPPG), or radar signals [18,31,68]. Jung et al. derive the HRV from the HR data [18]. RR is usually measured via PPG and BCG [15,57]. Some research uses magnetic and radar sensors [51,68]. Aiming at identifying the driver's stress level [15,27], four papers covered GSR signals [15,27,28,30]. Some research collects data on body movement and facial emotion by video or infrared cameras [43,56]. Using such data, the authors detected abnormal situations such as discomfort and drowsiness. The driving behavior may also potentially indicate mental health problems, such as cognitive ability, particularly for old people. Built-in sensors can provide the number of trips, driving duration and distance, time driving on local streets and highways, time driving during daylight, after dusk, or at night, and the number of hard breaks or accelerations [58]. The GPS device records environmental parameters like the driving areas [49].

Purposes
We found eleven papers toward solving health problems. However, most of them (n = 9) aim at improving driving safety. Stress and driving fatigue are detectable through HR and HRV [15,18,29,39]. EMG may also be incorporated as a source to extract features [42]. In order to detect drunk driving, the alcohol concentration is measured through a gas sensor [50], and driving behaviors under drunk conditions also could be a clue [31]. Several works explore the effectiveness of ECG for driver identification [37]. Discomfort detection benefits from monitoring body motion by an infrared camera and pressure pad [56].
Despite the popularity of research on developing ECG detection and HR measurement, we found no research using the measured data for solving clinical problems. Only two papers conducted strict subject selection (elderly people) and were aimed at Alzheimer's disease [49,58].

Discussion
In this paper, we analyzed up-to-date research on in-vehicle health monitoring using unobtrusive sensors, their locations, the biosignals recorded, and the (medical) purposes of recording. Based on our in-vehicle health monitoring terminology, we systematically reviewed the literature. Three databases performed our query and return almost a thousand responses. Interestingly, the records returned from PubMed and IEEE Xplore are entirely subsets of those from Scopus. This is in contrast to our findings in previously performed systematic reviews [69] and may be explained with the special focus on vehicles, which might not be covered completely in PubMed, as vital signs and medical applications are not in the focus of IEEE Xplore. However, the high sensitivity of our retrieval strategy is indicated by the fact that only less than 5% of the initially returned records survived our taxonomy-based standardized selection process and was included in this report (46/959 = 4, 8%).
We developed a graph (Figure 4) that is based on our terminology ( Figure 1) and our review results. It can serve as a framework or guide when developing in-vehicle health monitoring. A list of possible sensors as well as their suggested locations can be derived from the biomedical data that is needed for the respective purpose or application. Assuming the driver's heart parameters, such as HR, HRV, and RR, can be measured reliably while driving, a medical check-up could be done in the car on a daily basis during commuting. ECG-or BCG-based monitoring of these parameters enables early detection of heart disorders, such as atrial fibrillation [70,71], the precursor of stroke. To this end, the possible routes are highlighted with blue lines in Figure 4. Furthermore, we list the health purposes that can be supported by in-vehicle health monitoring, including heart diseases, respiratory diseases, epilepsy, psychological disorders, Parkinson's, and cognitive disorders. In addition, the annotated connections with references (Table 2) denote the weight of the connections. The connections with more supporting references suggest the promising approaches with higher feasibility. The table also provides clear linkages to the related references.
The graph in Figure 4 provides the key to answer our initial questions (Section 1): • Which sensors are suitable for in-vehicle data collection? Contact and capacitive electrodes capture ECG, and HR and HRV are computed from the recordings. Radar and magnetic induction sensors also are used for HR and RR measurements by detecting electron-magnetic signals due to organ movements, while BCG and piezoelectric sensors as well as accelerometers achieve similar goals through mechanical changes. Cameras provide video data, from which rPPG is generated. Furthermore, HR or HRV is extracted from the video data. Other work profiles driving behavior from vehicle built-in sensors in combination with GPS.

•
Where should the sensors be placed? Car seat and steering wheel host sensors that are in direct or indirect contact to the driver, e.g., capacitive and contact electrodes, respectively. The control panel and the windscreen are the suitable locations to mount video or infrared cameras.

•
Which biosignals or vital signs can be monitored in the vehicle? A variety of biosignals or vital signs are monitored already in the vehicle, including body-related (e.g., body temperature, EMG, GSR), heart-related (e.g., HR, ECG), blood-related (e.g., pulse transit time, oxygen saturation), lung-related (e.g., RR), and eye-related (e.g., saccade frequency) parameters. Some work focus on other information like driving behavior, gas concentration, emotion, body plethysmogram, grasping force, and body motion. • Which purposes can be supported with the health data? Driving requires intensive engagement in terms of both mental and physical efforts. The performance of driving is also associated with health problems, such as cognitive disorder [58,72]. As known, essential tremor is associated with incident dementia [73], and the monitoring of hand/foot tremor, for example, by detecting the operation of steering wheel or pedal gives us the possibility to assess the driving performance. The research in Stage I (Figure 3) points out the hidden clinical values of measuring biosignals while driving. However, most application-driven research (Stage II in Figure 3) is aimed at driving safety, such as fatigue detection and drunk driving. Though in-vehicle health assessment has potential in monitoring cognitive disorders, it is not yet developed to deliver medical monitoring in a clinical sense.
So far, a standardized terminology for in-vehicle health monitoring has not been established. For example, control panel is also called dashboard in many cases; GSR is also known as electrodermal activity (EDA). The proposed terminology (Figure 1) in this work, on the one hand, may promote the consistency in scientific communication and, on the other hand, it can form a structured description of a monitoring system. Surely the scientific community needs to further develop the terminology.
Driving makes the vehicle a dynamic environment. To improve the robustness of health-monitoring systems, one biosignal or vital sign can be measured by multiple sensors. For example, HR is obtained from ECG and PPG. Thus, multiple sensors yield a recording system with redundancy. Furthermore, when multiple parameters are intended to be observed, a body area network (BAN) is formed [74]. As a result, sensor coupling is an issue that must be considered for signal and data synchronization in future work.
Besides sensors, the vehicle is equipped with actuators [67]. Coupling such actors with health monitoring can perform signaling alarms in emergency cases, for example, when detecting heart attacks. Sensors and actuators communicate via the in-vehicle information system (IVIS) [75]. Schneider et al. transferred the capacitive ECG through the controller area network (CAN) bus [38]. Open issues include, but are not limited to, (1) how much of the large amount of continuously monitored data should be stored within the IVIS, (2) how the health data shall be managed, (3) how the data fusion workflow shall be implemented in the IVIS, and (4) data exchange between IVIS and health information systems (HIS).
The existing literature has shown the potential of unobtrusively monitoring health data in a mobile environment. The IoT sensors are nevertheless only able to measure physiological states but not interpret the underlying causes of them. Therefore, intrinsically linked mechanisms need to be put in place so that the collected data can be analyzed and interpreted by health professionals.
The reviewed literature focuses on health monitoring of the driver while passengers are rarely considered. We also presume that drivers will play a key role as their interactions with the vehicle provides relevant information. If we consider multiple occupants, the seating location provides a key to assign the data to the subject. However, it is a question of in-vehicle data storage and processing whether passengers shall be monitored at all. This will require instantaneous cloud transfer of data and cloud-based identity management. Then, scenarios can be satisfied where, for instance, a person sits in the driver seat and then moves to the passenger seat of a different car.
Digital technologies introduce ethical issues such as privacy vulnerabilities for users [76]. Accompanied with the vehicle telematics, physiological information tells far beyond the driving behavior. The driver can be identified with the physiological data [37]. The ownership and access to the data set are still open. Avoiding health information disclosure must be considered from the very beginning [77], and any link between the vehicle and external systems must be secured to protect from hacking health data.
To foster the car as a diagnostic space, strictly designed clinical studies need to be conducted. So far, the number of subjects is still low, and the significance of the effects of monitored data has not shown up. Semantic integration of sensor data for in-vehicle health management requires the establishment of standards [78].

Conclusions
As an equipped private space, the smart vehicle is a promising facility for health monitoring. This work provides a guide with regard to sensors, location, biosignals, and the purpose of currently proposed sensor systems for in-vehicle monitoring. Potential sensors can be derived from the biomedical data that is needed for the respective purpose or application. The suggested locations for the corresponding sensors are also linked. The annotated connections with references denote the weight of the connections and provide clear linkages to the related references. To the four sub-questions: (1) Unobtrusive sensors, which are based on electro-magnetical, mechanical, and optical mechanisms, are used for in-vehicle health data collection. (2) Driver-centered locations, such as steering wheel, car seat, and windscreen, are the most commonly used locations to host the sensors. (3) Typical physiological signals/parameters, such as ECG, HR, RR, body temperature, can be measured reliably, even while driving. (4) To date, most research has focused on sensor technology development. Research on mental health assessment and profiling the driving behavior is on track. However, health-oriented research on the medical use of physiological parameters is still on-demand. Furthermore, the terminology used in literature analysis may promote consistency in scientific communication and form a structured description of a monitoring system. Funding: This research received no external funding.

Conflicts of Interest:
The authors declare no conflict of interest.

Abbreviations
The following abbreviations are used in this manuscript: ("Document Title":biosignal OR "Document Title":biological OR "Document Title":biological signal OR "Document Title":physiological OR "Document Title":physiological signal OR "Document Title":physiological parameter OR "Document Title":vital signal OR "Document Title":vital sign OR "Document Title":vital parameter OR "Document Title":ECG OR "Document Title":electrocardiograph OR "Document Title":electrocardiography OR "Document Title":electrocardiogram OR "Document Title":heart rate OR "Document Title":heart rate variability OR "Document Title":heartbeat OR "Document Title":respiration rate OR "Document Title":respiratory rate OR "Document Title":breathing rate OR "Document Title":breathing OR "Document Title":breath OR "Document Title":respiration OR "Document Title":body motion OR "Document Title":driving profile OR "Document Title":routine OR "Document Title":driver's condition OR "Document Title":health state OR "Document Title":driver condition) AND ("Document Title":car OR "Document Title":car's OR "Document Title":vehicle OR "Document Title":in-vehicle OR "Document Title":in-car OR "Document Title":driver OR "Document Title":driver's OR "Document Title":driving OR "Document Title":automotive OR "Document Title":road OR "Document Title":safety belt OR "Document Title":steering wheel OR "Document Title":seat belt) Appendix A.3. Scopus TITLE ( ("biosignal" OR "biological" OR "biomonitoring" OR "biological signal" OR "physiological signal" OR "physiological parameter" OR "vital signal" OR "vital sign" OR "vital parameter" OR "ECG" OR "electrocardiograph" OR "electrocardiography" OR "electrocardiogram" OR "heart rate" OR "heart rate variability" OR "heartbeat" OR "respiration rate" OR "respiratory rate" OR "breathing rate" OR "breathing" OR "breath" OR "respiration" OR "body movements" OR "body motion" OR "driving profile" OR "routine" ) AND ( "car" OR "car's " OR "vehicle" OR "in-vehicle" OR "in-car" OR "driver" OR "driver's" OR "driving" OR "automotive" OR "road" OR "safety belt" OR "steering wheel" OR "seat belt" ) ) AND