Internet of Things-Based Smart Helmet with Accident Identification and Logistics Monitoring for Delivery Riders ^†

Abstract

The study developed a smart helmet prototype that prioritizes delivery rider safety and facilitates logistical communication for small businesses. This was achieved with a smart helmet, utilizing IoT equipped with crash detection and logistics monitoring functions. Various sensors such as an accelerometer and alcohol sensors were calibrated to improve accuracy and minimize errors. A mobile application was introduced to coordinate delivery logistics and track the location of drivers. The system had 90% accuracy in distinguishing real accidents, and it also had drunk driver detection with an accuracy of 88%. An ATTM336H GPS module was used for geolocation tracking, and a mobile application built with Bubble.io and Firebase was integrated into the helmet to send alerts the shop owners of Roger’s Top Silog House who provided delivery drivers as participants for the study, who gave us positive feedback indicating that our smart helmet performed very well and exceeded expectations.

1. Introduction

The use of commercial transportation, such as motorcycles, for the purpose of collecting, transporting, and delivering documents, parcels, and packages from various sectors (i.e., mail, food, and carrier) has been increasing; this has become a main driver in augmenting the essentials in the delivery industry [1]. However, motorcycle accidents are associated mostly with injuries and fatalities. The cause of these involve behavioral conditions, such as substance abuse, helmet wearing, violations [2], and even road environmental conditions [1]. According to a 2018 WHO Philippines report, 53% of the 11,264 road accident deaths were attributed to two-wheeled and three-wheeled riders and passengers, with 90% of them not wearing helmets. As of December 2020, there were 7,328,116 registered motorcycles, including 1,949,589 new units [3]. In 2018 alone, an average of 86 daily cases of motorcycle-related road crashes were recorded based on the annual report released by the MMARAS (Metro Manila Accident Recording and Analysis System) [4]. In 2021, there were 14,870 persons injured in motorcycle-related crashes, giving an average of 41 individuals per day [5]. In 2022, accidents involving motorcycles alone consisted of 22.59% of the total road accidents in Metro Manila, 31,124 of which were motorcycle accidents, with 17,089 of those resulting in injuries and 313 resulting in fatalities [6]. Even if road accidents are inevitable, humans involved in life-or-death situations rely heavily on the speed of emergency response. The study designs a smart helmet prototype that values the safety of the delivery rider and provides logistical information between the delivery rider and management. Using IOT-based solutions, the objectives of the study consist of the following: (1) design an IOT smart helmet accident detection system that gathers data from the accelerometer; (2) design a breath analyzer using MQ3 testing for the rider’s drunkenness during the smart helmet’s operation; (3) create a mobile application which notifies management of the delivery history and GPS location status; and (4) assess smart helmet’s operation quality upon usage by the delivery rider. The study only focuses on using a smart helmet with an embedded IoT system and Wi-Fi communication protocol. It enables data transmission through a Firebase IoT cloud server for the backend and utilizes Bubble.io for the frontend. This system notifies management about delivery riders, including crash detection and logistics information.

2. Methodology

This section contains an overview of the system’s block diagram as a prototype of an IoT-based smart helmet, with all of the relevant sensors, system design, as well as the processes behind its operation together with its circuit connection and implementation.

2.1. System Design

The smart helmet incorporates various sensors such as an accelerometer, vibration, alcohol and pressure resistive sensors, and GPS for point-to-point logistics tracking. Figure 1 shows the flow of operation. The data collected by these sensors are transmitted to the ESP32C3 microcontroller, which then relays it to Firebase as the backend database. This information is reflected in the Bubble.io mobile application for end-users, including administration (admins) who can monitor delivery manpower and potential accidents, delivery riders who use the app to accept deliveries and locate destinations, and customers who can track their deliveries. Figure 2 displays the wiring of the smart helmet.

2.2. System Architecture

This includes a calibration of the two (2) main sensory component systems for the creation of the smart helmet, a crash and alcohol detection system and a GPS system. The primary data that is determined here is sent to Firebase before it reflects any of those data onto the application. Figure 3 displays the flow of information of the accelerometer, vibration, and alcohol sensor. All data is sent to the MCU to be stored in Firebase. This data is interpreted and is displayed unto the mobile application interface.

Figure 4 shows the communication of the GPS module to the MCU. Data are sent to Firebase to undergo external API map routing in order to be properly displayed on the mobile app, which is then reflected, as shown in Figure 5 below.

2.3. Testing Procedure

The steps taken for the implementation of the smart helmet are as follows: (1) The admin and delivery rider create an account and connect the smart helmet via pocket Wi-Fi. (2) The delivery rider wears the helmet to begin the calibration of the accelerometer, vibration and alcohol sensors for initialization. (3) After approximately 2 min of warm-up time for the GPS, the admin can start placing orders for the delivery rider to accept. (4) After the order is accepted for delivery, the alcohol sensor samples the delivery rider’s breath. If they are determined to be sober, they can proceed with delivery. If not, the sensors can notify all admins that the delivery rider is drunk. (5) The smart helmet continuously sends all sensor data for potential accidents/crashes and intoxication detection, as well as tracks the location of the delivery using GPS. (6) Once the delivery is confirmed to be successful through the mobile app, the sensors would still run in the background to check for accidents, intoxication, and the location of the delivery rider, ready to notify admin of their logistics.

1.

Accelerometer Threshold

An accident is determined via the smart helmet whenever the accelerometer responds with x, y, and z values that go over the threshold of 12G or roughly 117.6 m/s² [7]. It then counterchecks with the vibration sensor’s output, determining whether the rider has been involved in an accident or not, as shown in Equation (1) as follows:

$$\left|\mathrm{a}\right|=\sqrt{\left({\mathrm{ax}}^2+{\mathrm{ay}}^2+{\mathrm{az}}^2\right)}$$

(1)

where |a| is the magnitude of linear acceleration and a = acceleration.

2.

Confusion or Error Matrix

To examine the reliability of the accelerometer and vibration sensor in accident detection, the following are calculated: the accuracy, as shown in Equation (2), the precision, as true instances of true positives shown in Equation (3), the recall, as the true positive rate shown in Equation (4), and the F1 score, as the harmonic mean between precision and the recall, shown in Equation (5), in terms of the error matrix of the results recorded from ten trials.

Accuracy = (TP + TN)/(TP + TN + FP + FN)

(2)

Precision = TP/(TP + FP)

(3)

Recall = TP/(TP + FN)

(4)

F1 Score = 2TP/(2TP + FP + FN)

(5)

where TP = true positive results, FP = false positive results, TN = true negative results and FN = false negative results.

3.

Calibration of MQ3 Sensor

To calibrate the MQ3 sensor as a breath analyzer, we extract the analog values from the sensor with a range of 0 to 4095 (with a resolution of 12 for the ESP32C3). R2 is taken from the physical resistor on the sensor, Ro (resistance of the sensor in normal conditions) is taken from the MQ3 datasheet, and Rs (output resistance of the sensor to alcohol) is used to acquire the blood alcohol content (BAC), computed using Equation (6). This value should be less than 0.05% to qualify as “sober”.

BAC = ab^(ratio)

(6)

where a is the BAC-intercept and b = slope.

4.

Root Mean Square Error

In order to assess the accuracy, the researchers compare the proposed system (smart helmet) over the conventional system (smartphone) and compute the difference between the resulting two values using Equation (7).

$$RMSE=\sqrt{\frac{{{\displaystyle \sum }}_{i=1}^n{\left(Conventiona{l}_i-Propose{d}_i\right)}^2}{n}}$$

(7)

5.

Time Delay, Likert Scale, and Standard Deviation

The time delay from the entirety of the data communication of the proposed system– from the embedded IoT device up to the data that are reflected upon the mobile application–is calculated using Equation (8). The questions are subdivided into three categories: (1) reliability, (2) usability, and (3) functionality, wherein responses follow a scale of 1 to 5, with 1 = strongly disagree and 5 = strongly agree, to undergo a Likert scale evaluation. The standard deviation can be used to provide additional information about the variability or consistency of the data, as shown in Equation (9).

Mean time delay = sum of trials/number of trials

(8)

$$\sigma =\sqrt{\frac{\Sigma {\left(x_i-\mu \right)}^2}{n-1}}$$

(9)

where x_i is the individual values from sample, µ = sample mean and n = sample size.

3. Results and Discussions

This section is divided to present the results for each objective of the testing and implementation of the prototype’s system to identify accidents and logistics monitoring.

3.1. Design of an IOT Smart Helmet Accident Detection System That Gathers Data from the Accelerometer

As detailed in

Table 1. Crash Detection from Accelerometer and Vibration Sensor Data.

No. of Trials	Situation of Test	Expected Output	Obtained Output	Interpretation
Trial 1	Helmet dropped 2 m above ground	T	T	Success (TP)
Trial 2	Helmet dropped 2 m above ground	T	T	Success (TP)
Trial 3	Helmet dropped 2 m above ground	T	T	Success (TP)
Trial 4	Helmet placed on the ground	F	F	Success (TP)
Trial 5	Helmet dropped 2 m above ground	T	T	Success (TP)
Trial 6	Helmet placed on the ground	F	T	Failed (FP)
Trial 7	Helmet dropped 2 m above ground	T	T	Success (TP)
Trial 8	Helmet dropped 2 m above ground	T	T	Success (TP)
Trial 9	Helmet placed on the ground	F	F	Success (TP)
Trial 10	Helmet dropped 2 m above ground	T	T	Success (TP)

, the crash detection was tested using an accelerometer and a vibration sensor that determined whether the impact was a crash or not attained. This was conducted over 10 trials for situations with expected results, and these were then compared with the obtained results to determine whether the smart helmet succeeded in detecting an accident.

As shown in

Table 2. Analysis of crash detection data.

Accuracy	Precision	Recall	F1 Score
0.90	0.875	1.0	0.933

, the accuracy of the system in detecting the situation as a crash yielded 90% while having a precision of 87.5%, a recall of 100%, and an F1 score of 93.3% from drop testing the smart helmet over 10 trials.

3.2. Design of a Breath Analyzer Using MQ3 Testing for the Rider’s Drunkenness during the Smart Helmet’s Operation

As shown in

Table 3. Analysis of alcohol detection.

Accuracy	Precision	Recall	F1 Score
0.890909	0.878787	0.960264	0.917721

, the accuracy of the MQ3 sensor to act as a breath analyzer had an accuracy of 89.09% while having a precision of 87.87%, a recall of 96.03%, and an F1 score of 91.78% from testing with different concentrations of alcohol readily available in the market.

3.3. GPS Location and Status of the Delivery

As shown in

Table 4. Comparison of longitude and latitude between GPS.

	GPS from Smart Helmet		GPS from Smartphone
No. of Trials	Longitude1	Latitude1	Longitude2	Latitude2
Trial 1	121.1922234	14.4668264	121.191716	14.4668255
Trial 2	121.1898427	14.4656059	121.1896703	14.4656555
Trial 3	121.1923470	14.4655427	121.1923521	14.4655419
Trial 4	121.1925436	14.4661518	121.1925606	14.4661125
Trial 5	121.1922318	14.465562	121.1922424	14.4655219
Trial 6	121.1936799	14.4655621	121.1936354	14.4655812
Trial 7	121.1908917	14.4654065	121.1908906	14.4653565
Trial 8	121.191133	14.4650616	121.1910309	14.4650906
Trial 9	121.19221	14.4645827	121.1923497	14.4645032
Trial 10	121.193271	14.4641768	121.1932287	14.4640736
RMSE = 0.000051274 (Latitude); 0.00017925 (Longitude)

, the lower the RMSE value, the better the fit. Based on the calculations, the RMSE of the latitude equates to 0.000051274. Meanwhile, the longitude is 0.00017925, implying that the smart helmet is on par with GPS from smartphones. Figure 6 shows the implantation of the smart helmet together with the GPS feature below.

As shown in

Table 5. Time delay from the helmet to application.

No. of Trials	Data from Helmet to Application (s)	Mean Time Delay (s)
Trial 1	3.764	2.37480108
Trial 2	1.449
Trial 3	1.66
Trial 4	2.024
Trial 5	3.915
Trial 6	1.256
Trial 7	2.185
Trial 8	6.657
Trial 9	3.059
Trial 10	1.423

, the mean time delay of the 10 trials is 2.37 s, which is considered real-time.

3.4. Assess the Smart Helmet’s Operation Quality upon Usage by the Delivery Rider

As shown in

Table 6. Weighted mean of responses using a 5-point Likert scale.

Category	Questions	Responses	Actual Mean	Standard Deviation	Interpretation
Reliability	5	5	4.24	0.469	Great
Usability	5	5	4.12	0.561	Great
Functionality	5	5	4.36	0.570	Great

, the actual mean and S.D. of the reliability of the smart helmet and mobile application are 4.24 and 0.469. The actual mean and S.D. of the usability of the smart helmet and mobile application are 4.12 and 0.561. The actual mean and S.D. of the functionality of the smart helmet and mobile application are 4.36 and 0.570. The smart helmet and mobile application performed greatly for the three categories. Figure 6 displays the actual helmet in use.