WO2022103336A1 - Methods and devices for determining tropical environment protective gear wearing - Google Patents

Abstract

[0085] Methods and devices for determining tropical environment protective gear wearing are provided. The device for monitoring protective gear wearing includes a first sensor, a second sensor and a processor. The first sensor is located internal to a protective gear and configured to sense microclimate humidity internal to the protective gear and the second sensor located external to the protective gear and configured to sense an ambient humidity external to the protective gear. The processor is coupled to the first sensor and the second sensor and configured to determine a humidity difference between the microclimate humidity and the ambient humidity and, in operation, to classify a state of the protective gear into a worn state and a not-worn state based on the humidity difference between the microclimate humidity and the ambient humidity.

Description

METHODS AND DEVICES FOR DETERMINING TROPICAL ENVIRONMENT PROTECTIVE GEAR WEARING

PRIORITY CLAIM

[0001] This application claims priority from Singapore Patent Application No. 10202011292X filed on 13 November 2020.

TECHNICAL FIELD

[0002] The present invention generally relates to protective gear such as industrial or military protective gear, and more particularly relates to methods and devices for tropical environment helmet wearing determination.

BACKGROUND OF THE DISCLOSURE

[0003] Appropriate helmet use offers the user mechanical protection to prevent workplace head related injuries and deaths. Here, the term “appropriate use” distinguishes itself from the naive compliance safety approach in having workers to wear a helmet at all times. Wearing a helmet at all times brings about thermal discomfort to the wearer as heat and humidity builds up around the head during use, affecting the user’s mental state and physical performance. This is exacerbated in tropical countries like Singapore and its South East Asian neighbors which experience outdoor temperatures of over 30 degrees Celsius with humidity of 60% to 90%. Users are compelled to remove their helmets regularly for relief which undoes the protection helmets are meant to provide. Thus, helmet research has focused on various cooling strategies and studies that measured users’ thermal comfort as it impacts their work performance. [0004] Although attractive to think the helmet’s appropriate use challenge as exclusively a challenge in thermal comfort and human performance, there are more factors hidden in anecdotes. The issue is further compounded by the fact that manufacturing sectors with significant outdoor activity like shipyards and construction industries require workers to put on protective masks or balaclava together with the helmet in place to, for example, prevent perspiration from entering the eyes to ensure workers can see properly and to block direct sunlight exposure as prolonged direct sunlight exposure may cause skin-related occupational disease. As such, workers have to ensure regular breaks for their well-being, while under a constant pressure to improve productivity.

[0005] However, when considered alongside workplace realities to stay profitable and the ever-present minority of protective headgear flouters, evidence of such safety dilemmas remains hidden and allows genuine work redesign suggestions or unsafe work grievances to go undetected. Singapore, as part of its forward-looking efforts, is targeting to implement best practices of workplace safety with worker safety ownership by 2028. Often considered to be the last-mile in total workplace safety, the worker safety ownership strategic thrust primarily consists of avenues for worker involvement in safety management, worker education and reporting systems for unsafe work conditions.

[0006] Thus, there is a need for methods and devices for protective headgear monitoring which overcome the drawbacks of the prior art and provides increased accuracy and worker comfort and, further, is particularly suitable for tropical climates. Furthermore, other desirable features and characteristics will become apparent from the subsequent detailed description and the appended claims, taken in conjunction with the accompanying drawings and this background of the disclosure. SUMMARY

[0007] According to at least one aspect of the present embodiments, a device for monitoring protective gear wearing is provided. The device for monitoring protective gear wearing includes a first sensor, a second sensor and a processor. The first sensor is located internal to a protective gear and configured to sense microclimate humidity internal to the protective gear and the second sensor located external to the protective gear and configured to sense an ambient humidity external to the protective gear. The processor is coupled to the first sensor and the second sensor and configured to determine a humidity difference between the microclimate humidity and the ambient humidity and, in operation, to classify a state of the protective gear into a worn state and a not-worn state based on the humidity difference between the microclimate humidity and the ambient humidity.

[0008] According to another aspect of the present embodiments, a protective gear wearing monitoring method is provided. The protective gear wearing monitoring method includes collecting calibration data and generating a calibrated threshold using a binary machine learning model to characterize the calibration data. The protective gear wearing monitoring method also includes monitoring wearing of protective gear using the binary machine learning model to determine whether the protective gear is worn or not worn in response to the calibrated threshold.

[0009] According to yet a further aspect of the present embodiments, a computer readable medium is provided. The computer readable medium includes instructions for a processor to perform a protective gear wearing monitoring method, the instructions causing the processor to collect calibration data and generate a calibrated threshold using a binary machine learning model to characterize the calibration data. The instructions also cause the processor monitor wearing of protective gear using the binary machine learning model to determine whether the protective gear is worn or not worn in response to the calibrated threshold.

BRIEF DESCRIPTION OF THE DRAWINGS

[0010] The accompanying figures, where like reference numerals refer to identical or functionally similar elements throughout the separate views and which together with the detailed description below are incorporated in and form part of the specification, serve to illustrate various embodiments and to explain various principles and advantages in accordance with present embodiments.

[0011] FIG. 1 depicts a schematic illustration of machine learning logistic algorithm in helmets in accordance with present embodiments.

[0012] FIG. 2, comprising FIGs. 2A to 2E, depicts photos of a Future Assault Shell Technology (FAST) style helmet utilized in accordance with the present embodiments, wherein FIG. 2 A a top, front, left perspective view, FIG. 2B depicts a rear view showing the datalogger and ambient sensor, FIG. 2C depicts an internal view with showing a location of a microclimate sensor location, FIG. 2D depicts the helmet on a mannequin head without a mask, and FIG. 2E depicts the helmet on the mannequin head with a mask.

[0013] FIG. 3, comprising FIGs. 3 A and 3B, depicts graphs of temperature versus time for a NO-MASK action sequence in accordance with the present embodiments, wherein FIG. 3A depicts a graph of a human subject and FIG. 3B depicts a graph of a mannequin experimental control.

[0014] FIG. 4, comprising FIGs. 4A and 4B, depicts graphs of per cent relative humidity versus time for a NO-MASK action sequence in accordance with the present embodiments, wherein FIG. 4 A depicts a graph of a human subject and FIG. 4B depicts a graph of a mannequin experimental control.

[0015] FIG. 5, comprising FIGs. 5A and 5B, depicts graphs of ambient-microclimate differences versus time for a NO-MASK action sequence in accordance with the present embodiments, wherein FIG. 5A depicts a graph of a human subject and FIG. 5B depicts a graph of a mannequin experimental control.

[0016] FIG. 6, comprising FIGs. 6A and 6B, depicts graphs of ambient-microclimate difference rate of change versus time for a NO-MASK action sequence in accordance with the present embodiments, wherein FIG. 6A depicts a graph of a human subject and FIG. 6B depicts a graph of a mannequin experimental control.

[0017] FIG. 7, comprising FIGs. 7A and 7B, depicts graphs of temperature versus time for a MASK action sequence (i.e., wearing a MASK as shown in FIG. 2E) in accordance with the present embodiments, wherein FIG. 7A depicts a graph of a human subject and FIG. 7B depicts a graph of a mannequin experimental control.

[0018] FIG. 8, comprising FIGs. 8A and 8B, depicts graphs of per cent relative humidity versus time for a MASK action sequence in accordance with the present embodiments, wherein FIG. 8A depicts a graph of a human subject and FIG. 8B depicts a graph of a mannequin experimental control.

[0019] FIG. 9, comprising FIGs. 9A and 9B, depicts graphs of ambient-microclimate differences versus time for a MASK action sequence in accordance with the present embodiments, wherein FIG. 9 A depicts a graph of a human subject and FIG. 9B depicts a graph of a mannequin experimental control.

[0020] FIG. 10, comprising FIGs. 10A and 10B, depicts graphs of ambientmicroclimate difference rate of change versus time for a MASK action sequence in accordance with the present embodiments, wherein FIG. 10A depicts a graph of a human subject and FIG. 10B depicts a graph of a mannequin experimental control.

[0021] FIG. 11 depicts a graph of an exemplary binary logistic regression (LR) model using ambient-microclimate humidity difference when a human subject is not wearing any mask in accordance with the present embodiments.

[0022] FIG. 12 depicts an illustration of a goodness-of-fit distribution of each dataset as a logistic regression model in accordance with the present embodiments.

[0023] FIG. 13 depicts an illustration of an efficacy distribution of each dataset as a logistic regression model in accordance with the present embodiments.

[0024] And FIG. 14 depicts an illustration of an efficacy distribution between subject and subject-control of each dataset in accordance with the present embodiments wherein for subject-control, a logistic regression model from a human subject was used to determine states using its corresponding run’s control data.

[0025] Skilled artisans will appreciate that elements in the figures are illustrated for simplicity and clarity and have not necessarily been depicted to scale.

DETAILED DESCRIPTION

[0026] The following detailed description is merely exemplary in nature and is not intended to limit the invention or the application and uses of the invention. Furthermore, there is no intention to be bound by any theory presented in the preceding background of the invention or the following detailed description. It is the intent of present embodiments to present methods and devices to determine whether a helmet or other protective gear is being worn by a human subject using machine learning algorithm(s) with low-cost, low-power sensors, specifically targeting microclimate characteristics within the protective gear, thereby providing a data-driven worker safety-ownership solution.

[0027] Although helmet researchers remain split with regards to various thermal comfort interventions, their raw temperature and humidity data consistently shows two distinct states between ambient and helmet microclimate locales during use, thereby enabling the present embodiments to leverage binary-output machine learning algorithms based on differences in ambient and helmet microclimates to characterize worn and not- worn states. Furthermore, the ability to model immediate data in accordance with the present embodiments overcomes the wrong determinations derived from setting fixed thresholds when used across different times and locations. The present embodiments present a cost-effective solution and appropriate form factor which can enable both commercial and user acceptance. This is in contrast to recent technologies such as image analytics which requires a large number of cameras and direct view of sight which is both costly and impractical at many worksites. Accordingly, the present embodiments facilitate data logging which will not only detect helmet misuse but also suggest that a helmet may have been over-prescribed for work, thereby providing an unbiased and reliable database for advantageously guiding work redesign towards appropriate helmet use to improve user mechanical protection and prevent workplace head related injuries and deaths.

[0028] In accordance with the present embodiments, a novel built-in algorithm with sensor fusion is provided to determine if a subject is actually putting on a helmet in any climate conditions without requiring skin contact. Additional sensors such as Global Positioning System (GPS) sensors and/or indoor trackers could also be provided to add information of localisation with geo-tagging. The algorithm provides systems, methods and devices in accordance with the present embodiments an essential foundation in data assurance to determine a subject is wearing a helmet using data analytics.

[0029] The present embodiments provides sensor integration to measure a helmet’s microclimate psychrometry with reference to an environment external to the helmet. When a helmet in accordance with the present embodiments is worn by a subject, the unique sensor fusion algorithm and design in accordance with the present embodiments will advantageously distinguish the helmet being worn.

[0030] In addition, the use of ambient sensors provides an integral design for a safety helmet in accordance with the present embodiments. The ambient sensors provide necessary information of the environment external to the helmet as reference for scientific study of the helmet’s microclimate psychrometry. Moreover, additional sensors like GPS sensors, indoor localisation sensors and fall sensors form an integral sensor suite for monitoring a subject’s location and well-being.

[0031] The methods, systems and helmets in accordance with the present embodiments provide advantageous solutions for industrial hardhat devices and attachments, such as hardhats utilized in the construction sector. In addition, the methods, systems and devices in accordance with the present embodiments can be utilized to provide training safety equipment for military and homeland security forces and for first responders, such as firefighters.

[0032] The methods, systems and devices in accordance with the present embodiments provide a form of safety assurance exceeding current solutions and technological benefits. Experiment runs involving human subject and mannequin experiment control were conducted across no-mask and mask conditions. There were four observations derived from helmet microclimate psychometry raw data logs from the sensors obtained during these experiments. First, only ambient- microclimate humidity difference (AMHD) was feasible. Second, the AMHD’s response was characteristically different between human subject and mannequin experiment control. Third, the rate of change of the AMHD exhibited high correlation to an exact moment when the helmet was worn or removed, giving near real time accurate information. Fourth, the temperature was invariant to ground truth in all cases.

[0033] The algorithm in accordance with the present embodiments models the parameter’s immediate field data into a binary logistic regression (LR) machine learning algorithm, characterizing not-wom and worn states with goodness of fit -80% for nomask condition and -75% for masked conditions. Tests to compare the AMHD LR model’s and respective ground truths demonstrated efficacies of -75% for no-mask and -70% for masked conditions. Further tests to compare the LR model’s and respective mannequin experiment control demonstrated the AMHD’s ability to identify flouting attempts. The results show that the algorithm can tell if a human subject is wearing the helmet or not. Thus, the present embodiments advantageously provide improved methods, systems and devices as compared to current state-of-the-art skin-contact and image analytics methods.

[0034] The present embodiments essentially comprise three elements: the algorithm; a hardware form factor and an installation method.

[0035] The algorithm, discussed in greater detail hereafter, can be further broken down into (i) an immediate calibration phase, (ii) a binary machine learning output model, and (iii) a normal use phase. The immediate calibration phase collects the data necessary for the later machine learning modelling and requires the helmet to be not- wom for a certain period (approximately one minute). Thereafter, the helmet is worn for the same duration to generate distinct sets of data as a base reference. The binary machine learning output model characterizes the data from the immediate calibration phase and generates an immediate calibrated threshold. The normal use phase is where the algorithm uses the model created to determine if the helmet is being worn or not.

[0036] The hardware form factor distinguishes itself from existing systems and devices in that it can be provided as a full-blown helmet or as devices which can be embedded into protective hard- shells. By design, the sensors’ hardware is an attachment completely separable from the helmet.

[0037] The installation method also distinguishes itself from conventional comparable installation methods. The hardware can be secured at the helmet’s back temporarily, such as by velcro, adhesive mountings or snap-on mechanisms. A microclimate sensor is positioned at the helmet’s interior, facing the wearer’s head. An ambient sensor is then positioned in the helmet’s exterior surface, ideally at the rear. Other optional add-on sensors at the rear include a GPS, one or more indoor tracking sensors, a fall sensor and other similar sensors to monitor the helmet wearer’s location and safety.

[0038] Referring to FIG. 1, a schematic illustration 100 depicts the operation of the machine learning logistic algorithm in helmets in accordance with present embodiments. The algorithm includes an immediate calibration phase 110, a binary machine learning modelling phase 120, and a normal use continuous monitoring phase 130.

[0039] The immediate calibration phase 110 includes a short predetermined timed calibration phase. As a non-limiting example, a two-minute minute calibration phase 110 can be split into one minute of readings 112 when the helmet is not worn and one minute of readings 114 while the helmet is worn. The readings 112, 114 are performed with adequate sensor sampling rates and reasonable times (e.g., one minute) determined in response to the eventual field application and providing statistically significant (e.g., N=30) microclimate readings for each of the two states. The modelling phase 120 classifies the helmet’s WORN and NOT-WORN states into an immediate-use binaryoutput machine learning model 125 using the baseline information from the calibration phase 110. Thereafter, the continuous monitoring phase 130 performs at work measurements 135 to process data received from the helmet and determine the helmet’s WORN and NOT-WORN states as well as logging other monitored conditions such as location or safety.

[0040] The advantageous key differentiators of the present embodiments are based on the algorithm with sensor fusion without the need for users to press “START” or “STOP” or perform numerous calibrations during the wearing process. The advantages of this algorithm are four as described hereinafter.

[0041] First, the methods, systems and devices in accordance with the present embodiments do not require skin contact to confirm a human wearer. In certain industries, some workers are required to wear face masks under helmets. It can be unimaginably hot for workers to operate under such working conditions. The present embodiments do not require direct skin contact of the skull to the helmet. Throughout testing of devices and systems in accordance with the present embodiments, consistent performance across no-mask and mask conditions was observed. These conditions cover construction workers habits in tropical climates of either directly wearing helmet on their heads or improvising with a towel for protection against sunlight and prevent perspiration from entering their eyes. In addition, as some conventional devices require the use of an electroencephalogram (EEG) to interface with the subject’s head, such devices will be subjected to various errors and false readings with every slight movement and placement changes. By not requiring skin contact, devices and systems in accordance with the present embodiments completely eliminates such data artefacts associated with EEG measurement.

[0042] Second, the methods, systems and devices in accordance with the present embodiments do not require personally identifiable or sensitive information. Although safety helmets are personalized items, the systems and devices in accordance with the present embodiments allow interoperability and sharing among teammates. Such requirement is necessary across the industries where safety management framework is compulsory. For example, in response to conventional personal protective gear developers’ tendency to lock customers in to their product by purposely developing products that are incompatible with personal protective gear of other developers, industries have established requirements for personal protective products to be compatible across different vendor solutions. In addition, one-person-only products are difficult to manage in the workplace, especially where work exigencies require equipment to be transferred or shared. Thus, the interoperable systems and devices in accordance with the present embodiments advantageously enable sharing among teammates with personal protective helmet solutions that are compatible across different helmets without diminishing the functionality of the present embodiments.

[0043] It follows from the no skin-contact requirement of the present embodiments, only using psychrometry relieves any requirement for a subject’s personal details. Unlike EEG signals, which are known to vary from person to person, the systems and devices in accordance with the present embodiments work for anyone who wears a helmet. Hence, the systems and devices in accordance with the present embodiments offer protection of personal identity and sensitive information by default. Moreover, the interoperability allows easy handover and usage across a whole factory floor or across construction teams without deliberate calibrations as required in conventional EEG-based devices.

[0044] Third, the methods, systems and devices in accordance with the present embodiments enable immediate field calibration, thereby overcoming the drawbacks of conventional thresholding methods. By design, the unique algorithm solution in accordance with the present embodiments includes prior-use immediate calibration 110 and can achieve very good continuous monitoring 130 efficacy via logistic regression machine learning 120 queued to the immediate environment. Experimental results show a 70% to 90% efficacy. Compared and contrasted with trite conventional methods requiring huge historical data, the present embodiments provide a neater, cleaner and more efficient solution which takes up less computer memory during actual operation. Another advantage of the methods, systems and devices in accordance with the present embodiments is the lack of a requirement for a fixed threshold. Conventional methods that use fixed thresholds, such as microclimate temperatures above 30° Celsius, to confirm a human wearer, are not be able to cover various operating conditions or adapt to multiple environments. In short, the methods, systems and devices in accordance with the present embodiments minimize the likelihood of false determination.

[0045] And fourth, the methods, systems and devices in accordance with the present embodiments use a humidity difference between ambient and micro-climate conditions to determine wearing status. The unique solution in accordance with the present embodiments depending on the accurate measurement of a humidity difference between ambient and a microclimate inside the helmet being worn by the subject provides an advantageous solution which is not dependent on the temperature, the humidity or other parameters of the helmet wearer’s environment. [0046] Referring to FIGs. 2 A to 2E, photographic images 200, 220, 240, 260, 280 depict a Future Assault Shell Technology (FAST) style helmet 210 utilized in accordance with the present embodiments. The image 200 is a top, front, left perspective view showing a datalogger 215 mounted on the helmet 210 in accordance with the present embodiments. The image 220 (FIG. 2B) depicts a rear view showing the datalogger 215 with ambient sensor. The data- logger 215 was attached to record temperature and humidity readings. No modification to the helmet 210 was needed as a Velcro system 225 is used to attach the data logger 215 at a back of the helmet 210. Two identical helmets 210 were used to facilitate experimental proceedings. Each data logger 215 was assembled with a battery powered Onion Omega 2 Plus single -board- computer (made by Onion Corporation of Boston, MA USA) accompanied by an Arduino-microcontroller docker accessory connected with a DHT-22 temperature and humidity sensor (made by Aosong Electronics Co Ltd, Guangzhou, China).

[0047] The image 240 (FIG. 2C) depicts an internal view of the helmet 210 showing a location of a microclimate sensor 242. The sensor 242 is a second DHT-22 temperature and humidity sensor and was placed within the core of the helmet 210 to register its microclimate, while the other sensor (i.e., with the data logger 215) was placed at the external surface of the helmet 210 to measure ambient environment. To eliminate erroneously repeated readings influencing the data’s statistical significance, the data loggers 215 are synchronised to record at time intervals coinciding with the DHT-22 sensors’ two-second measurement refresh rate.

[0048] The image 260 (FIG. 2D) depicts the helmet 210 on a mannequin 262 head without a mask. And the image 280 (FIG. 2E) depicts the helmet 210 on the mannequin

262 head with a mask 282. Two sets of thirty experimental runs were conducted to study the binary logistic regression (LR) machine learning algorithm’s accuracy and its central tendencies (i.e., for characterizing not-worn and worn states with goodness of fit). The first set was made without a mask as shown in the image 260 to represent the helmet 210 being worn directly on a user’s head, while the second set was performed with the mask 282 as shown in the image 280 to represent the user putting on a mask or balaclava. These experimental runs were conducted in room conditions and across non-consecutive days to ensure reproducibility under varying conditions in an indoor setting.

[0049] Each experimental run was an action sequence having a purpose of establishing a ground truth using the sensor readings. Ground truths, by definition, are information from direct observation. In this study, the ground truth is the actual state of the helmet 210 of either being worn or not. Establishing the ground truths enables the use of binaryoutput algorithms for further analysis. The action sequence consisted of a two-minute calibration period (i.e., the immediate calibration 110 (FIG. 1)), including a first minute in a NOT-WORN state, followed by a second minute in a WORN state. After the two- minute calibration period, the action sequence continued with a four-minute test period in which the helmet 210 was alternated between a one-minute NOT-WORN state and a one-minute WORN state. Both human subjects and mannequins (e.g., the mannequin 262 (FIGs. 2D and 2E) were subjected to the same abovementioned action sequence.

[0050] The head mannequin 262 was used as an experimental control to simulate reasonably sophisticated means to outsmart the device at the workplace. Two flouting scenarios were studied, the first being a total abandonment scenario of placing the helmet on a mannequin or an equivalent non-human device. The second case is based on the scenario where the acquainted user will misuse the helmet after going through the immediate calibration 110 procedures. [0051] Data was collected to analyse the modelling of the algorithm in accordance with the present embodiments and its resilience against potential flo ters . The analysis was made using Python’s sci-kit learn libraries and visualization of results was done using Excel standard plotter tools. All random sampling of the runs showed similar trends and two sets of data were picked for analysis. Typical NO-MASK and MASK runs were randomly selected to represent respective datasets in subsequent raw data log. The analysis of the data focused on application of the binary LR machine learning model using humidity and temperature parameters to determine the effectiveness and accuracy of LR algorithms in tropical settings and to determine which parameter would provide the best solution.

[0052] Referring to FIGs. 3A and 3B, graphs 300, 350 plot temperature, in degrees Centigrade, versus time (in two second intervals) for a NO-MASK action sequence in accordance with the present embodiments, wherein the graph 300 depicts a NO-MASK condition for a human subject wearing the helmet 210 and the graph 350 depicts a NO- MASK condition for the mannequin (i.e., the experimental control). The graph 300 depicts a ground truth 310 of a human subject (S) alternating between a WORN state 312 and a NOT-WORN state 314 while the microclimate temperature (MT-S) 320 inside the helmet is approximately equal to the ambient temperature (AT-S) 325 outside the helmet. The graph 350 depicts a ground truth 360 of the mannequin experimental control (C) alternating between a WORN state 362 and a NOT-WORN state 364 while the microclimate temperature (MT-C) 370 inside the helmet is approximately equal to the ambient temperature (AT-C) 375 outside the helmet with little variation between the WORN state 362 and a NOT-WORN state 364. From the graphs 300, 350, it is apparent that for the NO-MASK condition, temperature is not a responsive parameter. [0053] Referring to FIGs. 4A and 4B, graphs 400, 450 plot humidity, in per cent relative humidity, versus time (in two second intervals) for a NO-MASK action sequence in accordance with the present embodiments, wherein the graph 400 depicts a NO-MASK condition for a human subject and the graph 450 depicts a NO-MASK condition for the mannequin. The graph 400 depicts a ground truth 410 of a human subject alternating between a WORN state 412 and a NOT- WORN state 414. The microclimate relative humidity (MH-S) 420 inside the helmet varies as the helmet is worn and not worn, while the ambient relative humidity (AH-S) 425 outside the helmet does not vary. The graph 450 depicts a ground truth 460 of the mannequin experimental control alternating between a WORN state 462 and a NOT-WORN state 464. It can be seen from the graph 450 that the microclimate relative humidity (MH-C) 470 inside the helmet of the mannequin control remains approximately equal to the ambient relative humidity (AH-C) 475 outside the helmet as the helmet is worn and not worn. It can also be seen from the graph 400 that, for the NO-MASK condition, the microclimate humidity inside the helmet exhibits a response at a raw data level and the microclimate humidity response is due to a human wearing the helmet as the mannequin does not exhibit a similar response in the graph 450.

[0054] FIGs. 5A and 5B depict graphs 500, 550 which plot ambient-microclimate differences versus time (in two second intervals) for a NO-MASK action sequence in accordance with the present embodiments, wherein the graph 500 depicts a NO-MASK condition for a human subject and the graph 550 depicts a NO-MASK condition for the mannequin. The graph 500 depicts a ground truth 510 of a human subject alternating between a WORN state 512 and a NOT-WORN state 514. The ambient-microclimate humidity difference (AMHD-S) 520, in per cent relative humidity, between outside the helmet and inside the helmet varies as the helmet is worn and not worn, while the ambient-microclimate temperature difference (AMTD-S) 525, in degrees Centigrade, does not vary as the helmet is worn and not worn. The graph 550 depicts a ground truth 560 of the mannequin experimental control alternating between a WORN state 562 and a NOT-WORN state 564. It can be seen from the graph 550 that both the ambientmicroclimate humidity difference (AMHD-C) 570 between outside the helmet and inside the helmet and the ambient-microclimate temperature difference (AMTD-C) 575 between outside the helmet and inside the helmet do not vary as the helmet is worn and not worn by the mannequin control. Accordingly, it can be seen from the graph 500 that for the NO-MASK condition the ambient-microclimate humidity difference retains the microclimate humidity response (as seen in the graph 400) and that this response is due to a human wearing the helmet as the mannequin does not exhibit a similar response in the graph 550.

[0055] FIGs. 6A and 6B depict graphs 600, 650 which plot rate of changes in ambientmicroclimate differences versus time (in two second intervals) for a NO-MASK action sequence in accordance with the present embodiments, wherein the graph 600 depicts a NO-MASK condition for a human subject and the graph 650 depicts a NO-MASK condition for the mannequin. The graph 600 depicts a ground truth 610 of a human subject alternating between a WORN state 612 and a NOT-WORN state 614. The rate of change of the ambient-microclimate humidity difference (AMHDROC-S) 620, in per cent relative humidity per second, between outside the helmet and inside the helmet varies significantly as the helmet is worn and not worn, while the ambient-microclimate temperature difference rate of change (AMTDROC-S) 625, in degrees Centigrade per second, varies minimally as the helmet is worn and not worn. The graph 650 depicts a ground truth 660 of the mannequin experimental control alternating between a WORN state 662 and a NOT-WORN state 664. It can be seen from the graph 650 that both the ambient-microclimate humidity difference rate of change (AMHDROC-C) 670 and the ambient-microclimate temperature difference rate of change (AMTDROC-C) 675 are similar to each other and have little to no significant response. Thus, it can be seen from the graph 600 that for the NO-MASK condition the ambient-microclimate humidity difference rate of change (AMHDROC) exhibits an impulse response and that this impulse response is due to a human wearing the helmet as the mannequin does not exhibit a similar response in the graph 650.

[0056] Referring to FIGs. 7A and 7B, graphs 700, 750 plot temperature, in degrees Centigrade, versus time (in two second intervals) for a MASK action sequence (i.e., wearing a mask as shown in the image 280 (FIG. 2E)) in accordance with the present embodiments, wherein the graph 700 depicts a MASK condition for a human subject and the graph 750 depicts a MASK condition for the mannequin. In other words, the graphs 700, 750 are plotting similar parameters of temperature versus time as the graphs 300, 350 with the difference being that the graphs 300, 350 depict a NO-MASK action sequence while the graphs 700, 750 depict a MASK action sequence. The graph 700 depicts a ground truth 710 of a human subject alternating between a WORN state 712 and a NOT -WORN state 714. The microclimate temperature (MT-S) 720 inside the helmet and the ambient temperature (AT-S) 425 outside the helmet do not vary as the helmet is worn and not worn. The graph 750 depicts a ground truth 760 of the mannequin experimental control alternating between a WORN state 762 and a NOT- WORN state 764. The microclimate temperature (MT-C) 770 inside the helmet is approximately equal to the ambient temperature (AT-C) 775 outside the helmet with little variation between the WORN state 762 and a NOT-WORN state 764. From the graphs 700, 750, it is apparent that for the MASK condition, temperature is not a responsive parameter similar to the NO-MASK condition depicted in the temperature graphs 300, 350.

[0057] FIGs. 8A and 8B depict graphs 800, 850 which plot humidity, in per cent relative humidity, versus time (in two second intervals) for a MASK action sequence in accordance with the present embodiments, wherein the graph 800 depicts a MASK condition for a human subject and the graph 850 depicts a MASK condition for the mannequin. The graph 800 depicts a ground truth 810 of a human subject alternating between a WORN state 812 and a NOT- WORN state 814. The microclimate relative humidity (MH-S) 820 inside the helmet varies as the helmet is worn and not worn, while the ambient relative humidity (AH-S) 825 outside the helmet does not vary. The graph 850 depicts a ground truth 860 of the mannequin experimental control alternating between a WORN state 862 and a NOT-WORN state 864. It can be seen from the graph 850 that the microclimate relative humidity (MH-C) 870 inside the helmet of the mannequin control remains approximately equal to the ambient relative humidity (AH- C) 875 outside the helmet as the helmet is worn and not worn. Thus, the graph 800 indicates that for the MASK condition, similar to the NO-MASK condition of the graph 400, the microclimate humidity inside the helmet exhibits a response at a raw data level and the microclimate humidity response is due to a human wearing the helmet as the mannequin does not exhibit a similar response in the graph 850.

[0058] Referring to FIGs. 9A and 9B, graphs 900, 950 plot ambient-microclimate differences versus time (in two second intervals) for a MASK action sequence in accordance with the present embodiments, wherein the graph 900 depicts a MASK condition for a human subject and the graph 950 depicts a MASK condition for the mannequin. The graph 900 depicts a ground truth 910 of a human subject alternating between a WORN state 912 and a NOT-WORN state 914. The ambient-microclimate humidity difference (AMHD-S) 920, in per cent relative humidity, between outside the helmet and inside the helmet varies as the helmet is worn and not worn, while the ambient-microclimate temperature difference (AMTD-S) 925, in degrees Centigrade, does not vary as the helmet is worn and not worn. The graph 950 depicts a ground truth 960 of the mannequin experimental control alternating between a WORN state 962 and a NOT-WORN state 964. It can be seen from the graph 950 that both the ambientmicroclimate humidity difference (AMHD-C) 970 between outside the helmet and inside the helmet and the ambient-microclimate temperature difference (AMTD-C) 975 between outside the helmet and inside the helmet varies little as the helmet is worn and not worn by the mannequin control. Accordingly, it can be seen from the graph 900 that for the MASK condition the ambient-microclimate humidity difference retains the microclimate humidity response (as seen in the graph 900) and that this response is due to a human wearing the helmet as the mannequin does not exhibit a similar response in the graph 950. This is similar to the NO-MASK condition shown in the graphs 400, 450.

[0059] FIGs. 10A and 10B depict graphs 1000, 1050 which plot rate of changes in ambient-microclimate differences versus time (in two second intervals) for a MASK action sequence in accordance with the present embodiments, wherein the graph 1000 depicts a MASK condition for a human subject and the graph 1050 depicts a MASK condition for the mannequin. The graph 1000 depicts a ground truth 1010 of a human subject alternating between a WORN state 1012 and a NOT-WORN state 1014. The rate of change of the ambient-microclimate humidity difference (AMHDROC-S) 1020, in per cent relative humidity per second, between outside the helmet and inside the helmet varies significantly as the helmet is worn and not worn, while the ambientmicroclimate temperature difference rate of change (AMTDROC-S) 1025, in degrees Centigrade per second, varies minimally as the helmet is worn and not worn. The graph 1050 depicts a ground truth 1060 of the mannequin experimental control alternating between a WORN state 1062 and a NOT-WORN state 1064. It can be seen from the graph 1050 that both the ambient-microclimate humidity difference rate of change (AMHDROC-C) 1070 and the ambient-microclimate temperature difference rate of change (AMTDROC-C) 1075 vary similar to each other and have little to no significant response. So, it can be seen from the graph 1000 that for the MASK condition, the ambient-microclimate humidity difference rate of change (AMHDROC) exhibits an impulse response (similar to the NO-MASK condition AMHDROC response in the graph 600) and that this impulse response is due to a human wearing the helmet as the mannequin does not exhibit a similar response in the graph 1050.

[0060] As seen from the graphs hereinabove, microclimate humidity (MH) was found to be a more responsive parameter than microclimate temperature (MT) in detecting helmet-use for both with and without mask conditions, as shown from the results obtained under the two conditions (i.e., MASK and NO-MASK action sequences) with largely similar readings. This was clearly observable through the exponential growth and decay waveforms corresponding to human subj ect helmet’ s worn condition for both the NO-MASK and MASK action sequences as shown in the graph 400 (FIG. 4A) and the graph 800 (FIG. 8A), respectively. Although one can conclude that using microclimate humidity indicates the binary conditions of wearing and not wearing the helmet, a single humidity reading alone lacked ambient reference of environmental conditions to overcome reliability issues in single sensor designs.

[0061] Taking a difference between ambient humidity and helmet’s microclimate humidity as shown in the graph 500 (FIG. 5A) and the graph 900 (FIG. 9A) provided a better contrast as compared to conventional helmet monitoring systems and devices. The continued AMHD readings 520, 920 demonstrated preservation of exponential-like growth and decay waveforms characteristics, making binary LR modelling possible. This was possible as the ambient humidity remained fairly constant such that this difference could be sustained, thereby allowing use of the binary LR algorithm for helmet monitoring by reference to the ambient readings.

[0062] The subsequent rate of change 620, 1020 under AMHDROC as shown in the graphs 600 (FIG. 6A) and 1000 (FIG. 10A) matched with the above observations as it exhibited an inverse exponential growth curve in response to helmet use, followed by an exponential drop when the helmet was removed. The characteristics of these results provide strong evidence of an instance of helmet wearing condition or a helmet removal condition which is not dependent on whether a mask is worn or not worn.

[0063] In all experiments, a comparison was made with the control whereby the helmet was being put on and taken off the mannequin. In these control experiments, no significant change was observed from the raw readings, including the derivative data such as ambient-microclimate difference and its rate of change which exhibited similar invariance. Consequently, the LR models derived from these control experiments cannot distinguish if a human subject was using the helmet or not, since there were not distinct binary states in the data of the control experiments. This effectively confirmed the robustness of the proposed binary LR algorithm against potential attempts to misuse the helmet.

[0064] The ambient-microclimate humidity difference (AMHD) (see the graphs 500 (FIG. 5A) and 900 (FIG. 9A)) offers the best signal to noise ratio (SNR), making the use of low-cost sensors possible. The advantage of this derivative parameter overcomes the uncertainty presented by a single data point which often arises, while preserving microclimate response characteristics seen in the graphs 500 and 900. In addition, using AMHD is distinguished from conventional approaches that use multiple microclimate data points without any ambient information.

[0065] From analysis and anecdotal industry information, several salient adoption barriers for potential usage of commercially available helmet monitoring methods and systems were identified. First, sensor data from a helmet could be manipulated or doctored by a user, preventing any actionable intervention to be developed from data collected. Second, poor ergonomics due to weight distributions of various sensors and their associated batteries made the helmet difficult to be worn for prolonged periods and impeding safety of the helmet wearer. Thirdly, any modification to the mechanical structure of the protective helmet would require lengthy safety tests for legislation compliance. And lastly, ethical issues would arise as fringe developers begin to introduce forehead electroencephalography (EEG) functionalities to monitor attentiveness of helmet users. Such fringe development implicitly calls for involuntary creation and analysis of sensitive personal identifiable data, which undermines employer-employee relationships and trust, as it presents a significant privacy invasion to a common worker.

[0066] The data obtained from the various experiments exhibited in the foregoing discussion was tested by the Logistic Regression (LR) model. Referring to FIG. 11, a graph 1100 fits calibration datapoints 1110 onto the LR model 1120 as developed from the data of the graph 500 (FIG. 5A) in accordance with the present embodiments. The outlier datapoints 1110a, 1110b between -7 to -6 per cent relative humidity ambientmicroclimate humidity difference (AMHD) contributed to a goodness of fit ranging from 68.75 to 89.58% as shown at 1205 in the goodness of fit distribution illustration 1200 of FIG. 12. Comparing humidity based LR models 1210 with temperature based LR models 1220, the temperature LR models 1220 were considerably less or non-responsive with overall 60% goodness of fit across temperature related runs. Hence, temperature is not a viable parameter. These results were compiled and plotted out in the illustration 1200 to show the various accuracies under NO-MASK (N) and MASK (M) conditions. [0067] The AMHD parameter 1225 was the overall best parameter as shown at 1205. The LR models between human subject state 1230 and control state 1240 were distinguishable with the control 1240 states showing a considerable spread across 35.42 to 97.83% for a NO-MASK condition and convergence at 50% for a MASK condition. The ambient-microclimate humidity difference rate of change (AMHDROC) parameter 1250 was quantitatively the best parameter in fitting as a LR model, converging 87.5 to 97.92% goodness of fit. And there is a clear distinction between a human subject state 1252 and a control state 1254, which consolidated at 50.00%. The excellent 87.5 to 97.92% goodness of fit was due to small magnitude impulse response from the act of the helmet wearing being closer to a LR function, increasing goodness-of-fit as a result. However, it was found that the AMHD parameter 1225 is better than the AMHDROC 1250 against flouter scenarios and discussed hereinafter.

[0068] The LR algorithm in accordance with the present embodiments can advantageously easily pick out all control situations. The control data exhibits mostly 50% with the exception of ambient-microclimate humidity difference during NOMASK condition, showing a wide spread of accuracies ranging from 35.42% to 97.83%. This huge swing centered around the highly concentrated 50% accuracy with wide spreads stemmed from the invariant psychrometric changes. For cases not involving any human subject under control experiments, the LR model could only be half fitted or gives a wide spread. These observations enable differentiation between a human subject wearing the helmet versus a non-human subject wearing the helmet. In the latter case, the LR fittings at 50% or less gave clear indication against flouting attempts [0069] Efficacy is defined as the LR model’s ability to make correct determination during the continuous monitoring phase 130 (FIG. 1). Practically, the efficacy indicates parameter LR model performance when a user followed calibration instructions for the immediate calibration phase 110 (FIG. 1) with appropriate use of the helmet, where the immediate calibration phase 110 includes wearing the helmet and putting it aside.

[0070] Experimentally, continuous data was processed together with its corresponding LR model, with each outcome being compared against its ground truth as discussed hereinabove. There are two possible outcomes for each sensor’s reading, i.e., a correct prediction or a wrong prediction with reference to the LR model. Tallied results could be normalized on a 0 to 100% scale and a high accuracy would mean strong determination capabilities. The results are depicted in an efficacy distribution illustration 1300 of FIG. 13 which illustrate each parameter’s logistic regression model in making correct determinations against its run’s ground truth.

[0071] For similar reasons described above, microclimate-ambient humidity difference (AMHD) between 70 to 90% as shown at 1305 provides the best parameter in making the right determination through its LR model. Comparing humidity based LR models 1310 with temperature based LR models 1320, the temperature LR models 1320 were considerably less or non-responsive. These results were compiled and plotted out in the illustration 1300 to show the various accuracies under NO-MASK (N) and MASK (M) conditions. Even though a higher efficacy is desired, it is foreseeable that algorithm variations and improved hardware can provide improved efficacy. Moreover, the distinction of a human subject and the control group is clearly demonstrated with the efficacy and the LR model algorithm’ s responsiveness to a human subject. Temperature based parameters 1320 exhibited consolidation at 50% efficacy, which suggested that temperature was a poor parameter to be used in this case. [0072] Control groups continue to consolidate at 50%, indicating it would be difficult for flouters to fool the LR model. This would be represented in potential flouting scenarios whereby the flouter uses the helmet on a non-human head at all times.

[0073] Security is defined as the LR model’s ability to counter flouting attempts after proper calibration. This is different from the efficacies described above which follow prescribed calibration and normal use methods. Flouting attempts come in many forms, which in practice could be obeying a proper calibration scenario but followed by finding a reasonable substitute, like a mannequin head or a piece of rock to set the helmet on. The ability to distinguish between highly similar efficacies of genuine use against flouting attempts is necessary. It is noted that more sophisticated means of flouting such as using mannequin heads with skin-like texture or a sweat simulator are deemed too difficult to be executed at worksite.

[0074] In order to study flouting attempts, an existing LR model based on a human subject was benchmarked against the data obtained from its respective control experiment based on the mannequin. In practice, this was done on Excel programming and Python’s sci-kit to leam to process a mannequin’s normal-use data with the LR model created from a live human subject in accordance with the present embodiments. This arrangement is referred as “subject-control” group for its data cross application. The performance metric was similar to that developed for the efficacy illustration 1300, i.e., comparing the LR model with respect to ground truths with similar methods as previously described. Finally, the LR model outcomes with ground truths were normalized on a 0 to 100% scale and the results comparing subject and subject-control groups are shown in the illustration 1400 of FIG. 14. The illustration 1400 depicts an efficacy distribution between subject and subject-control of each dataset in accordance with the present embodiments wherein for subject-control, a logistic regression model from a human subject was used to determine states using its corresponding run’s control data.

[0075] Ambient-microclimate humidity difference (AMHD) 1410 showed a clear distinction between subject efficacies 1415 and subject-control efficacies 1420, indicating the LR model’s ability to distinguish flouting attempts in accordance with the present embodiments. The distinction is due to the field-calibrated LR model’ s decision threshold generated by the immediate subject. To elaborate, it has to begin with the raw data from the graph 500 (FIG. 5A), where the AMHD read -6% relative humidity at the start of calibration (e.g., the calibration phase 110 (FIG. 1)). Subsequently, the algorithm utilizes these data points to define the helmet off state 514 or the helmet on state 512 in the LR model, which in this case, the threshold was set at -6% relative humidity, as shown in the graph 1100 (FIG. 11). Then, using the control data as subject-control, which essentially ranged between 4% and 5% relative humidity in the graph 550 (FIG. 5B). Comparing this data to the -6% relative humidity threshold, the LR model would always output 1, indicating the helmet was always worn. Similarly, this would apply to masked condition in the graph 900 (FIG. 9), with a -12%RH threshold in this instance.

[0076] For all cases with control experiments, the LR model correctly quantify them at 50% efficacy. It gives a much tighter grouping for all the control experiments as seen in the illustration 1400 as compared to the illustration 1300 (FIG. 13). Therefore, it can be inferred from the LR model’s decision threshold and raw data plot on whether calibration procedures were adhered to by the human user. A low (~ 50%) efficacy strongly suggest a flouting attempt of helmet use with the help of a non-human object. [0077] AMHDROC 1430 also demonstrated clear distinction between subject efficacies 1435 and subject-control efficacies 1440 as compared to other parameters.

Although AMHDROC 1430 has its clear distinction, it is hard to rely only on this parameter solely. The reason is that AMHDROC as a parameter would not have a strong correlation of helmet wearing for extended periods of time. Its high performance in this study was due to the one-minute time windows prescribed in the experimental procedures. That is, whenever AMHDROC is close to plateauing back to zero, a helmet wear or removal would trigger a response from AMHDROC, as shown in both the graph 600 (FIG. 6A) and the graph 1000 (FIG. 10A). In such cases, the experiment’s procedure would result in an AMHDROC magnitude having high correlation to ground truth. However, in practice helmets are worn for more than a single minute’s duration and, as the helmet’s internal temperature and humidity build up, the ambientmicroclimate humidity difference would reasonably be expected to sustain at a nearconstant or slow-growth trend. In turn, the AMHDROC parameter would approach near- zero low-values or give rise to instantaneous spikes, as shown from the control’s data in the graph 650 (FIG. 6B) and the graph 1050 (FIG. 10B), which have no or low correlation to helmet wearing and removal.

[0078] Nonetheless, the AMHDROC’s impulse-like and directional response highly correlates to specific moments when the helmet is worn or removed. As seen from the raw data in the graphs 600 and 1000, the AMHDROC parameter exhibits a directional impulse-like response to helmet’s wearing and removal. When AMHDROC shows a positive impulse, it represents a helmet was just being worn while a negative impulse would represent removal. Thus, by monitoring the AMHDROC parameter, it is possible to automate the calibration process, making the process easy for users.

[0079] Thus, it can be seen that the present embodiments provide methods and devices for protective headgear monitoring having increased accuracy and efficacy while providing determination of user flouting attempts and which is particularly suitable for tropical climates where masks may or may not be used with the helmet. In accordance with the present embodiments, devices for determining tropical environment protective gear wearing include a first sensor internal to the protective gear, such as internal to a helmet, to measure the microclimate within including at least measuring the per cent relative humidity within the protective gear, and further includes a second sensor external to the protective gear to measure the ambient external environment parameters such as relative humidity. A processor coupled monitors the first and second sensors to measure at least an ambient-microclimate humidity difference to determine whether the protective gear is being worn.

[0080] The combination of first and second sensors, such as miniaturized sensors to eliminate a user’s discomfort, with the machine learning algorithm of the processor in accordance with the present embodiments, such as a binary logistic regression (LR) machine learning algorithm, advantageously enables determination of not-wom and worn states and has demonstrated the feasibility of devices and methods in accordance with the present embodiments. The binary LR algorithm in accordance with the present embodiments has successfully demonstrated that ambient-microclimate humidity difference (AMHD) is an attractive parameter to monitor helmets for appropriate use, applicable to both NO-MASK and MASK conditions. This is important for workplaces in tropical environments which makes skin contact wearable instrumentations extremely uncomfortable at work. As the binary LR algorithm in accordance with the present embodiments requires just a single parameter, the method in accordance with the present embodiments involves a simple immediate calibration step 110 (FIG. 1) to create the necessary binary output machine learning model, thereby enabling immediate application for processes such as continuous monitoring. By using low-power and low- cost miniaturized sensors with minimum sensor interfacing electronics, the devices in accordance with the present embodiments are suitable for long term application. [0081] The method in accordance with the present embodiments using the binary LR algorithm and AMHD parameter produced a model goodness-of-fit that converged at 80% (see 1205 in FIG. 12), making the model feasible in characterizing binary states; a converging 70% efficacy (see 1305 in FIG. 13) such that it is possible to deduce with reasonable confidence if the helmet is worn by observing a series of readings; and (3) the ability to identify flouting attempts (see the subject 1415 and the subject-control 1420 for the AMHD parameter 1410 in FIG. 14). Furthermore, the AMHDROC parameter 1430 provides impulse-like directional response to helmet wearing and helmet removal. When put together, the devices and methods in accordance with the present embodiments give a simple solution towards helmet wearing safety protocol.

[0082] Although AMHDROC has demonstrated higher performance quantitatively, it has its limitations and has to be combined with other parameters to provide the required solution. On the other hand, temperature has demonstrated slow response or invariant characteristics, making it unfeasible.

[0083] The above findings apply across NO-MASK and MASK conditions, which is possible for application in tropical environment where humidity can reach an uncomfortable level of 90%, rendering other forms of body contact wearable sensors unsuitable.

[0084] While exemplary embodiments have been presented in the foregoing detailed description of the present embodiments, it should be appreciated that a vast number of variations exist. It should further be appreciated that the exemplary embodiments are only examples, and are not intended to limit the scope, applicability, operation, or configuration of the invention in any way. Rather, the foregoing detailed description will provide those skilled in the art with a convenient road map for implementing exemplary embodiments of the invention, it being understood that various changes may be made in the function and arrangement of steps and method of operation described in the exemplary embodiments without departing from the scope of the invention as set forth in the appended claims.

Claims

CLAIMS What is claimed is:

1. A device for monitoring protective gear wearing comprising: a first sensor located internal to a protective gear and configured to sense microclimate humidity internal to the protective gear; a second sensor located on external to the protective gear and configured to sense an ambient humidity external to the protective gear; and a processor coupled to the first sensor and the second sensor and configured to determine a humidity difference between the microclimate humidity and the ambient humidity and, in operation, to classify a state of the protective gear into a worn state and a not-worn state based on the humidity difference between the microclimate humidity and the ambient humidity.

2. The device in accordance with Claim 1 wherein the processor is configured to utilize a logistic regression machine learning algorithm to classify the state of the protective gear into the worn state and the not-worn state based on the humidity difference between the microclimate humidity and the ambient humidity.

3. The device in accordance with Claim 2 wherein the processor is further configured to determine a calibrated threshold for the logistic regression machine learning algorithm to classify the state of the protective gear into the worn state and the not-worn state based on the humidity difference between the microclimate humidity and the ambient humidity.

33

4. The device in accordance with Claim 3 wherein the processor is further configured to automatically determine the calibrated threshold for the logistic regression machine learning algorithm in response to a rate of change of the humidity difference between the microclimate humidity and the ambient humidity.

5. The device in accordance with Claim 4 wherein the processor is further configured to determine user flouting attempts using the logistic regression machine learning algorithm in response one or both of the humidity difference between the microclimate humidity and the ambient humidity and the rate of change of the humidity difference between the microclimate humidity and the ambient humidity.

6. The device in accordance with any of the preceding claims wherein the second sensor is located on an external surface of the protective gear to sense the ambient humidity external to the protective gear.

7. The device in accordance with any of the preceding claims wherein the first sensor and the second sensor each comprise a miniaturized low power sensor.

8. The device in accordance with any of the preceding claims wherein the protective gear is a protective helmet.

9. The device in accordance with Claim 8 wherein the protective gear further comprises a mask worn under the helmet.

10. A protective gear wearing monitoring method comprising:

34 collecting calibration data; generating a calibrated threshold using a binary machine learning model to characterize the calibration data; and monitoring wearing of protective gear using the binary machine learning model to determine whether the protective gear is worn or not worn in response to the calibrated threshold.

11. The method in accordance with Claim 10 wherein the binary machine learning model comprises a binary logistic regression machine learning algorithm.

12. The method in accordance with Claim 10 or Claim 11 wherein collecting calibration data comprises: collecting calibration data of the protective gear not being worn for a predetermined time duration; and collecting calibration data of the protective gear being worn for the predetermined time duration.

13. The method in accordance with Claim 12 wherein the predetermined time duration is approximately one minute.

14. A computer readable medium comprising instructions for a processor to perform a protective gear wearing monitoring method, the instructions causing the processor to: collect calibration data; generate a calibrated threshold using a binary machine learning model to characterize the calibration data; and monitor wearing of protective gear using the binary machine learning model to determine whether the protective gear is worn or not worn in response to the calibrated threshold.

15. The computer readable medium in accordance with Claim 14 wherein the binary machine learning model comprises a binary logistic regression machine learning algorithm.

16. The computer readable medium in accordance with Claim 14 or Claim 15 wherein causing the processor to collect calibration data comprises causing the processor to: collect calibration data of the protective gear not being worn for a predetermined time duration; and collect calibration data of the protective gear being worn for the predetermined time duration.

17. The computer readable medium in accordance with Claim 16 wherein the predetermined time duration is approximately one minute.