An internet of things prototype to quantify the lateral wheel path distributions of local roads in New Zealand

ABSTRACT

The actual lateral wheel path distribution (LWD) on roads differs from the design assumptions. This leads to premature deterioration of chip-sealed pavements in the form of flushing and ravelling resulting in subsequent maintenance costs. The location of LWD is key to calibrating the variable-rate bitumen spray bar used to minimise flushing and ravelling during chip seal construction. Current methods used to estimate the location of LWD rely on visual inspection. This is subjective and reliant on individual expertise. This study aims to automate obtaining LWD by developing an Internet of Things (IoT) prototype to locate the LWDs. This prototype was tested by collecting LWDs from 22 locations in Christchurch, New Zealand. Results show that narrow roads exhibit a concentrated wheel path distribution. Specifically, increasing single lane carriageway width leads to a significant rise in the standard deviation (spread) and position of the LWDs among 22 locations investigated. Cycle lanes or parking bays significantly influence the location and distribution of LWD. There is no clear link between road camber and LWD. The findings can be used to improve the calibration of the variable-rate bitumen spray bar and inform pavement and traffic engineers to enhance construction and maintenance practices. .

KEYWORDS:

• Iot
• Wheel path distribution
• Single lane carriageway width
• Lane width
• Parking bay
• Cycle lane

1. Introduction

The motor vehicle fleet in New Zealand (NZ) has witnessed consistent growth over the past decade, reaching a staggering number of over 4.5 million vehicles in 2021 (Transport NZ 2023). However, there exists a lack of clarity regarding where these vehicles exactly drive on our roads. The prevailing assumptions in pavement design prescribe the construction of pavements with uniform layers based on the assumption that wheel paths are wide (Van der Walt et al. 2017a, van der Walt et al. 2017b). But it is well established that the deterioration of pavements predominantly occurs around wheel paths area, which signifies the importance of the distribution of traffic loads on the road (Zumrawi 2016, Clauß and Wellner 2022, Lindelöf et al. 2022). It is, therefore, imperative to enhance our comprehension of lateral wheel path distribution (LWD) to effectively predict and safeguard against premature pavement failure and improve design assumptions.

Most state highways and local roads in NZ are chip-sealed pavements (NZTA 2020). Chip-sealed pavements can supply a skid-resistant, water-resistant, cost-effective and hardwearing surface of the road. Chip-sealed pavements are non-structural pavements consisting of a layer of bitumen with embedded chips over a granular base, which generally needs resurfacing every seven to twenty years, depending on the traffic volume (NZTA 2005, NZTA 2023b). The New Zealand Transport Agency (NZTA) is intending to spend 4.2 billion dollars in the period of 2021–2024 on local maintenance of predominantly chip-sealed roads (Gransberg and Scheepbouwer 2010, NZTA 2023a).

One of the primary distresses observed in chip-sealed pavements is flushing (also known as bleeding) and ravelling (Rainsford et al. 2008, Karaşahin et al. 2016, van der Walt et al. 2017b). See Figure 1. Flushing is a pavement surface defect characterised by the upward displacement of the bituminous binder above the surfacing aggregate, often found within the wheel paths. Another prevalent defect is ravelling, which involves the loss of chips from the road surface. Ravelling typically occurs outside the wheel paths, around the centreline, between wheel paths, and on shoulders (Senadheera and Khan 2001, NZTA 2005, Patrick 2018).

Figure 1. Example of Flushing (also known as bleeding) and ravelling problems on chip-sealed pavements over a granular base in Christchurch, New Zealand.

The transverse variable bitumen spray bar is now commonly used to combat ravelling and flushing in NZ. By adjusting the application rate of bitumen across the width of the road, the occurrence of flushing and ravelling can be minimised (Pidwerbesky and Waters 2006). The variable spray bars apply a reduced amount of bitumen in the wheel path areas to control flushing and a higher application rate outside of the wheel path areas to mitigate ravelling (Pidwerbesky and Waters 2007, NZTA 2005, Pidwerbesky and Waters 2006, Rizzutto et al. 2015). However, a significant challenge associated with this technology is the absence of a well-defined design approach to determine the specific areas requiring high or low rates of applied bitumen (Rizzutto et al. 2015).

In NZ, Pidwerbesky and Waters (2007) compared pavement deterioration measures on roads constructed with variable and constant-rate bitumen spray bars in the Canterbury region. The main findings indicate that road segments built with a variable-rate spray bar showed a smaller reduction in macrotexture depth and binder rise than those constructed with a constant-rate spray bar (Pidwerbesky and Waters 2007). However, the optimal utilisation range of variable sprayers lacks clarity in the research, necessitating a more precise understanding of the distribution of vehicle wheel paths.

Multiple studies internationally have investigated the lateral placement of vehicles on roadways. The relationship between the lateral placement of vehicles and external variables such as rumble strips or bridges has been determined in studies. Porter et al. (2004) investigated the effect of centreline rumble strips using four tape switches set up in a geometric design. Their results showed that centreline rumble strips caused the traffic to move away from the centreline (Porter et al. 2004). Studies have investigated the effects of physical road features on LWD. Bowman and Brinkman (1988) investigated the effect of low-cost countermeasures (traffic calming), including combinations of advance warning signs, pavement markings, raised pavement markers, roadside delineators, type 3 object markers, and adhesive delineators on vehicle speed and lateral placement at narrow bridges using the FHWA Traffic Evaluation System. Their study found that these countermeasures significantly reduced speed variation and enhanced driving behaviour uniformity for all vehicle types and time periods (Bowman and Brinkman 1988). Similarly, King and Plummer (1973) investigated lateral placement and steering wheel reversals on a simulated bridge of different widths using a Greenfields Drivometre and video camera. Their results showed that people tend to drive closer to the centreline on narrower bridges (King and Plummer 1973). Das et al. (2016) investigated the effect different horizontal curves had on lateral vehicle placement using a video camera and marked chalk scale. The results showed vehicles moved more towards the centreline whilst taking a corner compared to straight roads. An increase in radius caused a further shift towards the centre of the road (Das et al. 2016).

Most of the studies were carried out internationally, and their findings are not comparable to NZ’s unique pavement and traffic conditions. Specifically, the lane widths are often very narrow, typically close to 2.7m (the lower limit) on two-lane, two-way roads in urban areas. Additionally, little or no nature strips or sidewalks exist in many areas (Auckland Transport 2021). The distance between the edge line and the drainage structure (curb and channel), is often very small in NZ. The vast majority of other studies were carried out on highways made from asphalt concrete and are not comparable to NZ chip-seal over granular base pavements. Furthermore, most studies have been conducted in right-hand driving regions rather than left-hand driving, the convention in NZ. None of the literature examined has investigated the effects of different road layouts on lateral vehicle placement. Most studies have relied on manual inspection of video footage or the use of expensive equipment.

Visual inspection of the variable-rate bitumen spray bars is still used to calibrate the variable-rate spray bar in terms of LWD, with many inherent issues. Visual clues are not always effective in identifying the precise failure mechanism (Patel et al. 2022). Flushing and ravelling, for example, can rapidly worsen, spreading across the entire pavement width (Gransberg and James 2005). Ideally, pavement resealing needs to be done before visible signs of distress appear, at which point in many cases, it’s too late to address cost-effectively (Burke 1998, Mamlouk and Zaniewski 1998, Gransberg and James 2005). The application is operator-dependent, and only experienced operators can manage the spray bar. Different inspectors may have varying expertise and experience, leading to inconsistent evaluations of the same pavement condition (Thodesen et al. 2012). Furthermore, as transportation networks continue to evolve rapidly, the manual collection of data has become burdensome for agencies like Waka Kotahi (NZTA) and contractors (Saleh and Van Der Walt 2019). Visual inspections are time-consuming and can be costly, particularly for large road networks (Koch and Brilakis 2011, Bianchini et al. 2010). LWD data collection by equipment is typically very expensive (ViaStrada Ltd 2008, van der Walt et al. 2017b). This may limit the frequency of inspections, leading to potential delays in identifying and addressing pavement issues in a timely manner. The traditional methods of collecting LWD often provide limited coverage. This can result in incomplete data sets that do not accurately represent the overall picture. The accuracy and variability of LWD can be affected by various factors. Road layouts, environmental conditions, variations in vehicle types and sizes, and the quality of data collection equipment can introduce inaccuracies and inconsistencies, reducing the reliability of the collected data.

Given these limitations, it is essential to complement visual inspections with more advanced and objective evaluation techniques, such as IoT, to obtain a more accurate and reliable assessment of the LWD on pavements. A data-driven approach is needed.

1.1. Objectives

With technological advancements, it is now possible to utilise affordable electronic components to construct ‘data collectors’. The Internet of Things (IoT) presents a promising solution for gathering data to address this issue discussed above. IoT involves deploying a network of small, cost-effective sensors for data collection, analysis, real-time data visualisation, and communication purposes (Abu-Elkheir et al. 2013, Hendricks 2015, Kim et al. 2017, Van der Walt et al. 2022). Van der Walt (2017) investigated LWD on highways using an expensive Infrared traffic logger (TIRTL) (van der Walt et al. 2017b). Li (2022) collected limited LWD data to determine the impact of parked or no parked cars on city roads based on an IoT prototype (Li et al. 2022). This study aims to extend the work by van der Walt (2017) and Li et al. (2022) and improve the information available to calibrate variable spray bars for local NZ conditions (van der Walt et al. 2017b, Li et al. 2022).

The objectives of this study are:

1. Develop a setup to assess the accuracy and usability of the US-026 distance sensor and IoT prototype.

2. Collect LWD distribution data on the four most typical road types using the IoT prototype in Christchurch, NZ.

3. Describe and compare the LWD data and make recommendations for further implementation and research.

This work aims to better understand the variations in lateral wheel path distribution in NZ conditions (data collected in Christchurch urban area). Obviously, it is not possible to address all the various permutations of road layouts, traffic factors, and other confounding variables and factors affecting LWD. Instead, this study will investigate four distinctive road layouts found in Christchurch, NZ, as a starting point as outlined above and use this to validate the use of the IoT prototype to collect LWD data and make further research recommendations.

This study contributes a specific, data-driven, comprehensive analysis of the abovementioned issues pertaining to chip-sealed roads, further enhancing the quality of chip-sealed pavements and extending the intervals between maintenance activities. The methodology employed in this study holds the potential to be applied to the collection of data for addressing other challenges encountered within the pavement domain.

2. Methodology

This methodology will present the configuration and development of the Arduino prototype. It will then describe how the prototype sensors are verified to be appropriate for data collection. The methodology will also discuss the accuracy of the sensors. Finally, the selection of data collection sites will be described.

2.1. Configuration and development of Arduino prototype

This paper uses the open-source Arduino ecosystem to set up an IoT prototype to collect LWD data. The total cost of the hardware was roughly $45 NZD. Arduino board is much more affordable than what has been used in previous studies. Previous work by van der Walt (2017) used expensive commercial-grade equipment called the Infrared traffic logger (TIRTL), which costs tens of thousands of dollars to buy and deploy commercially. Due to the lower cost, the Arduino microcontrollers have multiple advantages. For example, there is no need to personally supervise the instrumentation in case of theft. Various sensor units can be deployed simultaneously and connected to the network. Some disadvantages include weather tightness and finding appropriate places to mount the prototype, for example, a roadside pole.

2.2. How does the Arduino prototype work?

To classify and record the positions of passing vehicles, the US-026 ultrasonic distance sensors were used. Multiple prototypes were developed where sensors were connected to Arduino boards available in the university lab (‘Arduino Uno’ ∼$12 NZD each and ‘Arduino Nano 33 IoT’ ∼ $35 NZD each at the time of writing), which were combined with a real-time clock board (RTC) and a built-in SD card reader. Next, using the Arduino programming language based on C, code was uploaded to the boards. This code calculated the distance to the side panel of a vehicle by converting the time taken to send and receive ultrasonic waves, multiplying this by the speed of sound and dividing by two. The speed of sound at 15 °C was assumed to be 340 m/s. For efficiency of experimentation, this value was assumed constant for all experiments. While the constant assumption might introduce a slight variation, the difference in the speed of sound between 15 °C (340 m/s) and 20 °C (343 m/s),which were the temperature during data collection, does not significantly affect the outcomes of the experiments a difference of 0.00087%. Lastly, the code implemented a ‘loop’ to record passing vehicles, and the number of ultrasonic wave hits required to save a reading was set to five within a set margin of error. These values were determined through experimentation. At least five recordings within +/− 3 cm of each other were needed per vehicle to record the presence of a vehicle. The method was validated with video footage and distance markings on the road (This will need to be re-calibrated if experiments were to be conducted on roads with a speed limit of 100 km/h, not 50 km/h as fewer hits will register). The number of hits on a vehicle could also be used to loosely classify vehicles. For example, a bus or truck passing by the sensor would register many more hits than a traditional passenger vehicle. The average of at least five distances was recorded and saved as left wheel path distribution data detected by the sensor. As the data only represented the left-hand wheel path (the same as the TIRTL), the right-hand wheel path was modelled by adding an appropriate axle width and tire width, as discussed in Section 2.4 below.

2.3. Equipment verification

Collecting accurate distances to the outside wheel path with the US-026 sensor needed to be examined. Researchers were confident that the sensor could be reliably used to ‘hit’ a fast-moving vehicle and give reliable results. However, to verify the accuracy of the sensor and the prototype for this application, chalk lines spaced at 100mm intervals were positioned on the street, as shown in Figure 2. The chalk scale began from the road curb. A video camera was used to record where the left-hand front wheel of a vehicle hit the chalk scale. Still, frames taken from the recording were analysed with photo editing software to determine the distance between the outside edge of the wheel and the curb. Comparing multiple hits from the video information with that of the prototype allowed for the degree of error of the equipment to be calculated at different distances. The results of this analysis are shown in Figure 3.

Figure 2. The layout of the calibration experiment.

Figure 3. Schematic diagram of error rate for detecting distance changes on Creyke Road.

Creyke Road was chosen as the test site because it demonstrated ideal conditions and a safe environment for the researcher to carry out work safely (under the directions of the Health, Safety and Wellbeing plan). Comparing the sensor-recorded data and visual camera measurement (chalk lines on the road), the sensor’s accuracy was assessed. A bar plot showing how the recordings compared to the measured values is shown in Figure 4.

Figure 4. Bar plot comparison of recorded and measured distances (note that the right LWP is modelled. See method below).

Figure 3 shows that there is a tendency for accuracy to decrease along with distance changes. The farther the distance, the higher the error rate. In each test, the percentage difference between recorded and measured values was calculated, as shown in Figure 3. The percentage differences value is no more than 1.72%, equating to a difference of 4 cm. Thus, the accuracy of sensors and prototype is assessed to be reliable up to ∼ 3m. Additional measurements were made up to 5m with an error of 6%. However, very few readings (<10) were recorded at this distance.

2.4. Modelling the right wheel path – the inside wheel path distribution

The IoT prototype can only reliably establish the distance from the curb to the outside wheels of a vehicle. However, it is advantageous to model the inside wheel path, as well as it is helpful for calibrating the variable bitumen-spray bar, and it gives an indication of where the vehicles travel relative to the centreline of the road. To model the inside wheel path an approximation for axle width (centre to centre of tyre) was obtained. Table 1 shows 20 axle widths obtained by measuring vehicles directly and the centre-to-centre distance between the areas of flushing on 20 regional roads in Christchurch. The averages of both sample groups are given.

Table 1. Left: Separation determined through observation of wheel paths and measurements between areas of flushing on the roads of Christchurch. Right: Axle widths of common car varieties.

Road	Approx. Flushed area separation (cm)	Common Make, Model of vehicles in Christchurch	Body Shape	Axle Width (cm)
Shirley Road Eastbound lane	151	Mitsubishi Pajero	Truck	155
Ilam Road Northbound lane	148	Mazda CX5	SUV	162
Ilam Road Southbound lane	146	Honda CRV	SUV	155
Clyde Road Northbound lane	145	Subaru Impreza	Hatchback	149
Riccarton Road Eastbound lane	152	Mazda Demio	Hatchback	146
Waimari Road Southbound lane	145	Nissan Wingroad	Station Wagon	146
Creyke Road Westbound lane	154	Nissan Sentra	Station Wagon	143
Avonhead Road Northbound lane	151	Toyota Gaia	Mini Van	145
Maidstone Road Westbound lane	154	Mercedes-Benz	Sedan	154
Grahams Road Westbound lane	151	Hyundai Getz	Hatchback	145
Grahams Road Northbound lane	153	Toyota Corolla	Sedan	150
Peer Street Southbound lane	153	Mitsubishi Lancer	Sedan	152
Middleton Road Southbound lane	151	Ford Ranger	Ute	162
Hansons Lane Northbound lane	145	VW Golf	Hatchback	148
Wharenui Road Northbound lane	152	Holden Commodore	Sedan	153
Wharenui Road Southbound lane	145	Subaru Legacy	Station Wagon	145
Matipo Street Northbound lane	151	Mazda 6	Hatchback	148
Kotare Street Westbound lane	150	Toyota HiAce	Van	151
Straven Road Northbound lane	152	Hyundai i30	Hatchback	150
Withells Road Southbound lane	151	Mazda BT-50	Ute	161
Average	150	Average	151

The assumption of regional roads accommodating predominately passenger vehicles was confirmed via visual observation in the field. Thus, the axle width was taken as the average of the two samples in Table 1, which was 151 cm (with a standard deviation of 4.56 cm).

2.5. Site descriptions and selection

Twenty-two sites located in nine different districts were chosen for the investigation to give a representative sample across Christchurch. The nine stars in Figure 5 show the location of the suburbs where data was collected that provides a representative sample of the Christchurch city area. The IoT prototype was deployed at each of the 22 sites to collect LWD data. While data was being collected, researchers, based on visual inspection, took notes with location and time stamps, noting down interesting driver behaviour.

Figure 5. Distribution of data collection sites in Christchurch (Google Maps,2023).

A two-lane, two-way road is a single roadway with one lane travelling in each direction, which is typical for roads in Christchurch. Two-lane two-way roads in Christchurch can be further subdivided into four different road types, as shown in Figure 6.

Figure 6. Four different layouts of roads commonly found in Christchurch.

As shown in Figure 6, Type 1 roads are fully marked roads with parking bay and no cycle lanes. Type 1 is the most common road layout in Christchurch. Type 2 are roads without parking lines and no cycle lanes, but vehicles can park along the roadside. Type 3 roads are fully marked roads with parking bays and cycle lanes. Type 4 roads are marked roads with cycle lanes. The roads examined for each type are outlined in Table 2.

Table 2. Comparison of recorded results obtained from the four road layout types.

Types	Sample location	Suburb	Single lane carriageway width (m)	Lane width (m)	Parking bay width (m)	Cycle lane width (m)	Camber of road	No. Vehicles	Mean of left lateral wheel path (cm)	Normalised mean to 6.5 m road width (cm)	Standard deviation (cm)	75th percentile (cm)
Type 1	Sawyer arm road-H	Northcote	6	4	2	NA	Medium (∼4%)	148	369.49	394.49	22.06	381.25
Type 1	Barrington street-N	Spreydon	6.5	4.5	2	NA	Low (∼1%)	126	410.02	410.02	39.75	440.50
Type 1	Greers road-A	Bryndwr	6.5	4.5	2	NA	Medium (∼4%)	201	420.00	420.00	22.91	420.00
Type 1	Greers road-B	Bishopdale	6.5	4.5	2	NA	Medium (∼4%)	262	427.71	427.71	23.93	443.00
Type 1	Waimari road-J	Ilam	6.5	4.5	2	NA	Medium (∼4%)	101	411.62	411.62	27.26	426.00
Type 1	Springs road-P	Hornby	7	5	2	NA	Medium (∼4%)	78	431.46	406.46	44.45	467.75
Type 1	Innes road-F	Mairehau	7	5	2	NA	Low (∼1%)	75	465.33	440.33	29.66	484.50
Type 2	Waimairi road-K	Ilam	5.5	5.5	NA	NA	Low (∼2%)	183	427.98	477.98	31.12	450.00
Type 2	Sawyer arm road-I	Northcote	6	6	NA	NA	Medium (∼4%)	120	380.39	405.39	31.96	400.00
Type 2	Greers road-C	Bryndwr	6.5	6.5	NA	NA	Medium (∼4%)	50	411.72	411.72	25.25	425.00
Type 3	Waimari road-L	Ilam	6.5	3.2	1.8	1.5	Low (∼2%)	55	450.53	450.53	40.66	487.00
Type 3	Barrington street-O	Spreydon	6.5	3	2	1.5	Low (∼2%)	114	435.89	435.89	51.01	474.00
Type 3	Greers road-D	Bishopdale	6.5	3.2	1.8	1.5	Medium (∼4%)	317	397.35	397.35	23.81	412.00
Type 3	Springs road-Q	Hornby	7	3.5	2	1.5	Medium (∼4%)	76	446.29	421.29	28.57	467.00
Type 3	Ferry road-U	Woolston	7	3.2	2	1.8	Medium (∼3%)	182	449.25	424.25	32.09	470.75
Type 3	Kotare street-S	Fendalton	7	3.5	2	1.5	Medium (∼3%)	66	401.23	376.23	30.03	423.75
Type 3	Innes road-G	Mairehau	7	3.5	2	1.5	Medium (∼4%)	76	447.87	422.87	37.08	474.00
Type 4	Greers road-E	Bishopdale	4.5	3	NA	1.5	Medium (∼4%)	62	170.50	270.50	19.85	162.75
Type 4	Waimairi road-M	Ilam	5	3.5	NA	1.5	Medium (∼4%)	91	208.98	283.98	21.69	228.50
Type 4	Kotare street-T	Fendalton	5	3.5	NA	1.5	Low (∼2%)	69	238.5	313.50	21.06	251.50
Type 4	Ferry road-V	Woolston	6	4.5	NA	1.5	Low (∼2%)	62	182.85	207.85	44.78	210.00
Type 4	Springs road-R	Hornby	6.5	4.5	NA	2	Medium (∼4%)	50	202.06	202.06	46.57	236.75

Several variables must be considered, such as lane width, road segment, speed limit, vehicle type, and road layouts. In Christchurch, apart from highways and the central business districts, most roads have a speed limit of 50 km/h. The following variables were kept consistent for all sites. (1) all selected locations exhibited a straight road segment of at least 150m; (2) all roads were within a 50 km/h speed limit zone. (3) As reasonable as practical, recording locations were as far away as possible from intersections and other major road hazards. (4) The presence of researchers on the road was minimised as much as possible to avoid impacting driver behaviour. Two key variables, namely lane width and single lane carriageway width, differed significantly within urban Christchurch for all four road types. Additionally, researchers sought roads in Christchurch that exhibited all 1–4 layout types along their length. Thus, keeping other confounding variables as consistent as practicable. These sites (for example Greers Road discussed in Section 4.3) could then also be used to conduct a preliminary investigation into the transition between road Types 1–4 on the same road, meaning that most other variables are consistent and similar as practicable. Also, the road camber data also were collected by electronic level on site.

3. Results

3.1 . Preliminary data for four different road types

A summary of the results recorded from the 22 site locations is given in Table 2. The mean position distance of the left-hand wheel path from the left-hand road curb has been calculated for each sample. A normalised value for this mean position was also calculated to compare different widths. Figure 7 provides a summary of the metrics used to display results for this study in Table 2.

Figure 7. Schematic diagram of used terms in the paper.

Figure 7 depicts the layout of a lane, spanning from the road curb to the centreline (one side of a road, a single lane). It shows the typical location of a parking bay, a cycle lane, and one vehicle lane. Most cycle lanes in Christchurch are 1.5m wide. Two depictions of distribution are shown, the left wheel path (LWP) and right wheel path (RWP). It is important to note that lane width and single lane carriageway width are distinct measurements (not the same thing). single lane carriageway width is the full road width and encompasses various components, such as lane width, cycle lane width, and the width of a parking bay. The symbol μ represents the mean position of LWD, which is measured from the curb. The standard deviation, indicated by σ, represents the standard deviation of the LWD data. RWP data is modelled based on the axle width, which is defined as the distance between the centreline of two wheels on the same axle. The 75th percentile, also known as the ‘third quartile’, represents the 75th percentile of values. Both these measures can be used to calibrate the variable bitumen spray bar. Table 2 shows the results obtained from the IoT prototype on the 22 sites selected in Christchurch, NZ.

4. Analysis and discussion

In the case of Type 1 roads, Sawyer Arm Road-H exhibited single lane carriageway width of 6m, while Barrington Street-N, Greers Road-A, Greers Road-B and Waimari Road-J had single lane carriageway width of 6.5m. Springs Road-P and Innes Road-F had a larger single lane carriageway width of 7m. The mean position distances to the left-hand wheel path for these road locations were recorded as 369.49, 410.02, 420.00, 427.71, 411.62, 431.46, and 465.33 cm, as shown in Table 2. Correspondingly, the standard deviations for these measurements were 22.06, 39.75, 22.91, 23.93, 27.26, 44.45, and 29.66 cm. The 75th percentile were 381.25, 440.50, 420.00, 443.00, 426.00, 467.75 and 484.50 cm. Notably, this indicates an approximate upward trend in the mean position of LWD, standard deviation, and the 75th percentile as the single lane carriageway width or lane widths increase. This trend is likely attributable to the enhanced manoeuvrability and increased space available to drivers on wider roads, facilitating greater movement within the given road space.

Type 3 roads demonstrate a relatively high mean position of LWD. This characteristic can be attributed to the typical design of Type 3 roads, which includes provisions for parking and cycling lanes. These provisions result in a ‘pushing’ of vehicles inward to the centre from the curb.

Type 4 roads are typically comprised of a cycle lane alongside a vehicle lane. When examining the relationship between single lane carriageway width and the mean position of LWD, there is no consistent pattern observed. However, it is interesting to note that the standard deviation of LWD (spread) tends to increase with the single lane carriageway width or the lane width. It was found that it is not uncommon for drivers to move in and out of cycle lanes when there is no cyclist present. For example, in Type 4 roads, the cycle lane is 1.5m-2m wide from the curb, and often the LWD mean position is within or close to one standard deviation away from the cycle lane edge. This means that a significant number of vehicles encroach on the cycle lane. It should be noted that roads featuring adjacent cycle lanes tend to exhibit significantly narrower lane widths, primarily to accommodate the recent implementation of cycle lanes on existing roads in Christchurch, thereby reducing the lane width (NZTA 2015). Consequently, drivers may find themselves positioned closer to the shoulder line, as their intention is to maintain a central position. It was found that drivers still tend to perceive the cycle lane as a significant component of the drivable road width, as researchers found it not uncommon for vehicles to be inside the cycle lane. Given the narrow lane widths and in an effort to steer clear of the centreline and oncoming cars (hard barrier), drivers tend to travel in closer proximity to and occasionally encroach into adjacent cycle lanes when no cyclist is present. Results recorded indicate that drivers prioritise oncoming traffic compared to cycle lanes when they are unoccupied by cyclists and therefore tend to maintain a further separation from oncoming cars than the empty cycle lane. This behaviour needs to be further investigated. Cyclists and cyclists’ safety investigations were outside the scope of this research.

Results show that drivers on Type 1 roads opt to position themselves at greater distances from parked vehicles. Such behaviour may be attributed to the perception of parked vehicles as ‘a very hard barrier’, owing to their size and proximity to the traffic lane. To mitigate the risk of collision with these parked vehicles and a potential door-opening scenario, drivers may choose to navigate slightly closer to the centreline, thus creating a slight safety buffer between themselves and the parked vehicles.

Visually inspecting the raw data, there is a clear difference between the road layouts of ‘with parking bay and cycle lane (Type 3)’ and the roads ‘only with cycle lane (Type 4)’. Given the same lane width and cycle lane width, the mean position and 75th percentile and standard deviation of LWD of roads ‘with parking bay and cycle lane (Type 3)’ is greater than that of roads ‘only with cycle lane (Type 4)’. This discrepancy could be attributed to the feeling of a larger road overall (Type 3), thus having an increased comfort to manoeuvre within the road. Results and observations indicate that the empty cycle lane may provide a perceived buffer to the parked cars as the ‘hard barrier’. This result needs further investigation.

4.1. Probability density functions (PDF) for LWD

Figure 8 illustrates the overlap of Probability Density Functions (PDF) for 22 sites. Each line in the plot represents a different location.

Figure 8. PDF for the four different road layouts considered. a) is Type 1 b) is Type 2 c) is Type 3 and d) is Type 4.

For seven roads in Type1 in Figure 8 (a), a discernible trend emerges where the standard deviation of LWD consistently increases with the single lane carriageway width or lane width. Figure 8 (b) shows that Waimari Road-K is significantly different from the other two roads. The reason for this difference is that the road’s centreline is not well defined in this instance. This is further discussed in Section 4.4. In Type 3 roads, the PDF plots resemble those of Type 1. The mean LWD position values are consistently above 370 cm. The mean position and 75th percentile values of LWD for Type 3 roads surpass those of Type 1 roads overall. But the standard deviation of Type 3 varies significantly. Type 4 roads exhibit a distinction between Springs Road-R and Ferry Road-V compared to other roads in Type 4. Both roads have similar mean position and standard deviation values, possibly due to their identical lane widths of 4.5m. The remaining three roads, which have lane widths below 3.5m, exhibit a similar narrower standard deviation.

4.2. Linear interpretation of the LWD data

Figure 9 shows the overall analysis of the LWD data, investigating the relationships between LWD position (mean) and distribution (StD). In all the plots within Figure 9, a positive correlation is observed between single lane carriageway width, lane width, and LWD position. This implies that as single lane carriageway width or lane width increases, the LWD mean position or StD also increases. However, the relationship between lane width and LWD StD is much less pronounced compared to the other three plots.

Figure 9. The linear interpretation of the data. a) the linear interpretation between lane width and LWD mean position b) the linear interpretation between single lane carriageway width and LWD StD c) the linear interpretation between lane width and LWD mean position d) the linear interpretation between single lane carriageway width and LWD StD.

Table 3 was derived from the data presented in Figure 9 and Table 2. The details are shown as follows.

Table 3. Result analysis of the linear interpretation of the data.

	Lane width & Mean	Lane width & StD	Single lane carriageway width & Mean	Single lane carriageway width & StD
Equation	Y = 16.09X + 305.2	Y = 0.3639X + 30.1	Y = 98.40X −245.1	Y = 5.712X–4.212
R squared	0.02324	0.001328	0.5154	0.1940
P value	0.4982	0.8721	0.0002	0.0402
Deviation from zero?	Not significant	Not significant	Significant	Significant

The relatively modest R-squared values observed in the relationships between vehicle positioning and lane width or carriageway width, as 0.02324 and 0.001328 for ‘Lane width & Mean’ and ‘Lane width & StD’ are observed. This signifies that the linear regression models provide only limited explanatory power. These values suggest that variations in lane width or carriageway width alone do not significantly account for the observed fluctuations in vehicle positioning. The comparatively higher R-squared values for ‘Carriageway width & Mean’ (0.5154) and ‘Carriageway width & StD’ (0.1940) suggest an improved alignment between carriageway width and vehicle positioning metrics. This indicates that single lane carriageway width has a more discernible influence on these aspects measured. In sum, the overall R-squared values indicate that the chosen linear models only sum what capture the complexities underlying the relationships in the data recorded, thereby necessitating a bigger and more diverse sample of data is required to understand underlying complexities.

The calculated P values for the association between single lane carriageway width and the mean position and StD variables are less than the threshold of 0.05. This finding indicates a statistically significant relationship between single lane carriageway width and the mean position and StD variables exists. Specifically, as single lane carriageway width increases, there is a notable increase in the mean position (location) and standard deviation (spread) of LWD. Conversely, the data shows that the association between lane width and the mean position and StD variables is not statistically significant in the data collected. Both these findings should be further investigated in future research (a much larger data set) to understand the many integrated factors that affect drivers’ behaviour when positioning their vehicles on the road. From the results, there is no apparent relationship between the camber of the road and LWD data, suggesting that a normal degree of road camber does not significantly impact the lateral positioning of vehicles.

4.3. Case study – greers road

Greers Road exhibited all four-road road layout types (1-4 in Figure 6) along its length. Noting from Table 2 that Greers Road showed that the normalised mean LWD reduced from 420 to 270.5 cm going from road layout Type 1–4 examples shown below in Figure 10 (a–d). In the case of Greers Road, the segment with a single lane carriageway width of 4.5m (Greers Road-E) exhibits a mean of LWD measuring 170.50 cm and a 75th percentile value of 162.75 cm. The segment with a wider single lane carriageway width h of 6.5m (Greers Road-D) demonstrates a substantially higher 75th percentile value of 412 cm.

Figure 10. Greers Road displays all four types of road layouts along its length. Top left a) Type 1, top right b) Type 2, bottom left c) Type 3 and bottom right d) Type 4 (Google Street View, 2023).

The analysis conducted in Section 4.2 is supported by the observations and recorded data of Greers Road. It was specifically observed that for Greers Road, as the lane widths increased, the standard deviation of the LWD also increased.

4.4. Vehicles travel past the middle of the road

Figure 11 visually presents the lateral positions of the wheel path, encompassing both the left and right wheel paths, for Type 2 road layouts. The dashed line positioned on the leftmost side (lateral position 0) signifies the curb, while the rightmost line (green) line represents the centreline of Greers Road-C, with a single lane carriageway width of 6.5m. Similarly, the other dashed lines (orange and blue) correspond to the centreline of Sayers Arm Road-I, with a single lane carriageway width of 6m, and the centreline of Waimari Road-K, with a single lane carriageway width of 5.5m.

Figure 11. The distribution of LWD of Type 2 roads.

It is evident from Figure 11 that the LWD mean position values exceed the threshold of the centre line – in other words, cars often travel over the middle of the road. Notably, this is particularly pronounced in the case of Waimari Road-K, where vehicles tend to drive down almost the middle of the road where oncoming traffic is minimal, see Figure 12.

Figure 12. Image of a section of Waimari Road (Type 2) with no centre line, here vehicles tended to drive down almost the centre of the road.

4.5. Advantages and disadvantages of the prototype (IoT device) to collect LWD

Results show that there is a complex interplay of factors specific to each road that makes it challenging to establish a straightforward relationship for LWD. It is recommended that multiple factors mentioned in this paper be further investigated in follow-up studies.

It has been suggested that the number of IoT devices will reach a trillion units, with related profits exceeding $200 billion (New Zealand IoT Alliance 2018, Sundmaeker et al. 2010, Jalali et al. 2019, Pocock 2019, Rana et al. 2019). Based on the experiences developing this IoT prototype for LWD data collection, the advantages and disadvantages of this approach have been summarised in Table 4.

Table 4. Advantages and Disadvantages of the Arduino prototype (IoT device) to collect LWD.

Advantages

Disadvantages

- Very low cost - Easy to deploy/mount, no need to cross the road or interfere with traffic - Data-driven approach - Automated vehicle collection - Low power demand (3.3–5 V) - High sensor accuracy within 4m - Reduced workforce allocation - Real-time monitoring (timely) - Decrease paperwork - Smaller and more convenient equipment to procure - Adaptable to collecting other forms of data, for example, vehicle classification/ counting, is easily possible

- Calibration of the sensor unit can be a challenge, requiring many iterations - The current setup only collects the nearest wheel to the curb Must estimate the inside wheel path with a model - Sensors and electronics subject to water/ bad weather - Communication connections can be unreliable - Need to provide external power - The US-026 sensor accuracy severely drops past 4 m - Technology (data fusion and compatibility) - Limited high-skill workforce to interpret the data - Lack of relative design codes and regulations - Disputes over data ownership - Reliability and maintenance of many devices in the field (replacing one problem for another)

5. Conclusion

This paper has developed an IoT prototype that can locate the lateral wheel path distribution on single lane carriageways with two-way two-lane roads very cost-efficiently. This IoT prototype has been deployed at 22 sites to demonstrate the capability on local roads in Christchurch. The following key conclusions can be drawn from the results and analysis:

1. The IoT prototype was successfully deployed to collect LWD measurements on urban roads common to Christchurch. The US-026 sensor used for this application exhibits a measurement error of no more than 1.67% in the range of 2 cm–300 cm, corresponding to a difference of ∼4 cm. This suggests that the sensor provides sufficient accurate and reliable LWD measurements of moving vehicles within the specified range and context described in this paper.

2. The prototype and data-driven distribution information presented here can be used to help move the industry away from the reliance on visual inspection and can be used to better understand driver behaviour.

3. Results show the single lane carriageway width exhibits a statistically significant relationship with the LWD mean position and LWD standard deviation (spread), showing a notable increase as single lane carriageway width increases. The association between lane width and the LWD mean position, or LWD standard deviation, is not statistically significant. Further research is needed to understand the interplay of the many factors that impact driver behaviour.

4. The presence of adjacently parked cars significantly influences the LWD. There is a propensity for LWD values to be positioned toward the centreline, drivers avoiding ‘hard barriers’ such as parked cars. Cycleways, with no cyclist present, are perceived as a ‘softer barrier’. It was not uncommon for vehicles to move into the cycle lane.

5. No relationship was observed between the camber of the road and LWD, suggesting that road camber does not significantly impact the lateral positioning of vehicles in normal ranges up to 4%.

6. Despite its drawbacks, the usage of IoT prototypes developed here for the collection of LWD should be further investigated and extended with additional features added.

7. Further recommendations for the work have also been discussed below.

6. Limitations

Although efforts were made to address confounding variables through careful site selection and the collection of multiple samples, it remains challenging to eliminate external variables that may have influenced the behaviour of drivers in the study, as a live road is a dynamic environment. One such variable is the presence of the researcher wearing a high-visibility vest, which may have inadvertently influenced driver responses. Additionally, the effect of the Arduino devices was not investigated. The range of distance sensor detection for moving vehicles operating within a speed limit of 50 km/h is constrained to a maximum of 4–5m; therefore, it is only realisable and helpful for single-lane analysis. Moreover, the work was limited to tangent sections, and the behaviour of drivers on curves was not examined. While this paper has presented some general trends in Christchurch, NZ, it has not fully investigated the underlying reasons governing the LWD and human behaviour. It is important to expand the scope of investigation by examining a more significant number of sample locations 100 + and collaborating with social scientists to better understand human behaviour. This broader sampling approach would provide a more comprehensive understanding of the patterns and trends associated with LWD when the source is available for deployment of an IoT- network of IoT devices.

7. Recommendations for future work

It is recommended that this work could be extended in the following ways.

1. Conduct long-term studies to investigate the impact of data-driven variable bitumen spray bar calibration vs visual inspection.

2. The low power constraints allow this device to be powered by solar and deployed for long-term data collection in remote locations.

3. Extend this IoT prototype’s features to count and classify vehicles (heavy – light) accurately in multiple locations at a fraction of the cost of traditional methods. This can then be used to estimate Equivalent Standard Axles.

4. The low implementation cost of this prototype allows it to be used to investigate the LWD around bends by deploying multiple sensing devices simultaneously around the bends in different scenarios.

5. Investigate long-term LWD movement due to variables that change next to the road over time. For example, temporary traffic management scenarios could be an interesting area of cross-disciplinary work.

6. Combine this IoT prototype with other sensors to get multiple sources of truth for example rain sensors.

7. Extend this work to the state highway network and 100 km/h stretches of road (the impact of the Doppler effect on accuracy needs to be investigated).

8. Investigate the deployment of the IoT prototype with large IoT networks (not just Wi-Fi) in collaboration with local governments and telecommunication companies to develop a comprehensive IoT- sensor mesh across a city.

Disclosure statement

No potential conflict of interest was reported by the author(s).