-A major problem with vehicles used in densely populated areas is parking. Part of the problem is the scarcity of the resource in busy areas and the non optimal utilization of private spots. We propose the 'Smart Shared Private Parking' model that aims at including in the parking pool also private parking spots. In an urban area with both commercial and residential buildings, frequently household owners do not utilize their own parking lots during office hours, long duration of shopping and vacations. At the same time, such owners want to be able to park in their lots when needed. Therefore, to engage the residents and allow them to provide their parking spaces when unused one needs a real-time, distributed infrastructure that supports dynamic allocation of spaces. We propose a conceptual deployment of edge hardware and communication mechanisms to enable shared private parking sharing. We illustrate an infrastructure setup required to analyse various aspects of the model and show its effectiveness in parking allocation in terms of vehicular emission, utilisation rate, and revenue as metrics considering all the stakeholders. A simulation based on a district of Dublin provides quantitative instances for these metrics. Index Terms-V2X, VANET, Vehicular Communication, Smart Shared Private Parking, Smart Gate, Edge Computing
Urbanization and Industrialization are the major factors
contributing to the increased dependence on vehicles [
A novel model for the sharing of private parking lots that considers user preferences while offering the possibility to optimize utilization rate.
A design and implementation of the model.
A simulation based on the city of Dublin to evaluate the model effectiveness.
The rest of the paper is organized as follows. Section II overviews related work that addresses the problem of parking optimization and mobility models. Our proposed design is illustrated in Section III and its implementation in Section IV. The evaluation of the proposal by means of simulations is presented in Section V. A discussion and final remarks are offered in Section VI.
Innovative parking and mobility models have been
investigated before, e.g., Some of the works that are used to derive
the points of common knowledge are [
Modelling the parking demand and supply management has
been the goal of several works. The importance of different
parking modelling strategies is highlighted in [
1) Shared parking model I - Focus on overtime parking
information: The boons of shared parking and managing the
overtime stay through different correlation analyses are
illustrated in [
2) Shared parking model II- Focus on walking distance
and other preferences: Driver and owner preferences are the
focus of [
Large scale mobility models and simulation scenarios are necessary in assessing the routine traffic and formulating steps required to manage varying traffic scenarios. These models also serve as a base to assess different concepts of mobility such as parking and intermodal mobility.
1) MoST and PyPML: Simulation scenarios based on the
city-state Monaco are proposed in [
PyPML is a Parking Monitoring Library designed using
SUMO and using the Monaco SUMO Traffic (MoST) scenario
[
2) Smart parking using Vehicular Ad Hoc Network
(VANET) in the Dublin scenario: The authors of [
In addition to research work, a number of companies are
already providing services in the field of smart parking. Zenpark
is a shared parking solution available in France and Belgium
[
The work proposed employs smart gate hardware (with
the same capabilities as an RSU) in place of a common
RSU. The objective is to reduce the costs incurred due to
the provisioning of separate hardware units that generally add
up the expenses of a smart city. Conceptually, the hardware
is said to be in place of an electrical garage device to sense
the spots, communicate and actuate the mechanical parts of
the parking lot when required. In the work, each RSU or
smart gate hardware represent a parking lot in the simulation.
In the requirement analysis phase, we distributed a survey to
34 employees from various German locations of the company
Itemis AG. The survey was conducted in December 2020. The
goal was to assess the mindset of car drivers and residential
parking lot owners. The results of the survey are in line with
the the conclusions of the works [
The mathematical model of the work is designed
considering the concept of overtime parking and drivers’ preference to
walk for a certain distance. Utilisation rate Um in Equation 1
is the result of matching time preferences and additional
preferences of a vehicle with a parking lot. In a parking lot
m, the variable xmn (satisfying time preferences and additional
preferences) equal to 1 multiplied by duration of parking gives
the utilisation rate of parking lot m by a vehicle n. The sum
of utilisation by vehicles n1....nN at a particular hour gives
the overall utilisation of a parking lot m at a particular hour.
Equation 1 is an extended version of the mathematical model
in [
N n
P tdur xmn Um = n=1
Tm(H) (1) xmn = 81 if, (Dm > > > <
Dmax;n) & (Pm = Pn) &
n (tstart;m < t + tdur < tend;m) & >> (Fm > 0) & (t tlr;m > tth) >:0 Otherwise where Um is the utilisation rate of the parking lot ’m’, tdurn represents the duration for which the vehicle ’n’ needs parking. Tm(H) = tend,m-tstart,m represents the total duration of availability of a parking lot. tstart,m is the time at which parking lot ’m’ starts to being offered. tend,m is the time at which parking lot ’m’ is no longer available. But in the work, the value of Tm(H) is equal to 1 as we consider the simulation scenarios on a per hour basis. lr,m is the time at which a parking lot is last reserved. tth is the threshold time before which a parking lot can be utilized. Dm is the actual walking distance between the parking lot ’m’ and the destination. Dmax,n is the preferred walking distance by the driver of vehicle ’n’ to the destination. Pm is the set of service offering parameters from the residential parking lot ’m’ and Pn is the set of services demanded by the driver of vehicle ’n’. Fm is the number of free spots available in the parking lot ’m’. m = m1,.......mM, Each of the values of m represent the response i.e. by a parking owner offering a parking. n = n1,.......nN, Each of the values of n represent the demand i.e. a driver of vehicle ’n’ looking for a parking.
In Equation 3, Revenue Rm is the collective sum of parking charges and overtime parking charges incurred by vehicles n1,.......nN using the parking lot m. The decisions xmn and xod.mn (given by Equations 2 and 4) coordinate the decisions of considering the parking usage of vehicles whose timings and preferences have been matched with the parking lot.
N X[(tdnur n=1 xod:mn = Rm = xmn cm) + (tond
xod:mn cod:m)] (1 if, (t > tdnur) & (xmn = 1)
0 Otherwise where Rm is the revenue generated by a shared residential parking lot ’m’. tdurn is the duration for which the vehicle ’n’ occupies a parking lot. xmn is the decision variable that denotes whether a vehicle ’n’ occupies a parking lot ’m’ or not. cm is the fixed cost to occupy a spot in parking lot ’m’ for an hour. todn is the duration for which the vehicle ’n’ is parked overtime. xod.mn is the decision variable that denotes whether a vehicle ’n’ occupies a parking lot ’m’ overtime or not. cod.m is the cost to occupy a spot in parking lot m overtime for an hour.
The base framework used to implement the proposed model
is Eclipse MOSAIC [
Simulation of Urban Mobility (SUMO) is a free,
opensource simulation software for modelling inter-modal traffic
systems [
The Scenario-Convert tool is a part of Eclipse MOSAIC
[
The real demand is used in the simulation. We do so by resorting to the demand data from the official website of Dublin administration [10]. The data is collected by means of a survey taken at 33 locations around the city centre. The survey locations are strategically planned so that the number of inbound and outbound vehicles are tracked accurately. The survey locations are marked in Figure 3. We consider the number of vehicles moving in and out of the city at entry point 31. Routes denote the paths taken by the vehicles in SUMO network. Although the scenario-convert tool of Eclipse MOSAIC is capable of converting a set of geographic coordinates into SUMO routes, the newly generated routes cannot be directly merged into the route navigation database which is a small shortcoming of the tool (observed until December 2020). To overcome this, a custom script is developed to use the existing route generation functionalities of Eclipse MOSAIC.
The components involved in the simulation are vehicles and RSU. We developed applications that need to be deployed in each component. The application represents the behaviour of the component (say parking functionality in a vehicle). The modules that a vehicle in simulation possesses are a navigation module, an in-built operating system that coordinates with internal components and a communication module capable of Vehicle to Everything (V2X) communication. The vehicle’s application in the work proposed involves the usage of these three modules. RSU forwards the request from the vehicle to cloud-based web services. In the shared parking model, RSU represents the smart gate hardware which can be equipped with a vehicular communication stack, enabling the sensing of the arrival of vehicles and also the actuation of the gate, when required. The smart gate hardware is thus the edge component that ultimately enables the smart sharing of the parking lots. Similar to a vehicle, RSU is also equipped with a navigation module, a communication module and an Operating System (OS).
To help understanding the operation of the system, let us consider the two most relevant data flows, that is, the parking request and the resume signal.
1) Parking request: The parking request from a vehicle
flows through RSU to the web service. The vehicle or car
driver initiates the demand for a parking request by sending a
geocast message consisting of a request for parking. Geocast
message communication is a form of message communication
where the message is sent to all the potential receivers in a
geographical location [
2) Resume signal: The resume signal originates from a
vehicle when the parking is completed. The parking duration
varies randomly. In order to resume the journey, the resume
signal is sent to a RSU nearby, which forwards the signal
with all necessary details to the cloud. The aim of this signal
and shared residential parking lots. An important objective
is to match the vehicle’s preferred walking distance and the
distance between the parking lot and the destination of the
vehicle. Open Source Routing Machine (OSRM) provides
routing and navigation facilities offline and is based on OSM
data [
Then comes the matching strategies in which one of them matches the current time and duration of parking with the availability of parking lots. Another strategy is to match the preferences of vehicle with the services offered by a parking lot. Finally, an allocation module is present to distribute the demand. In a demand-supply management application, the important focus is to distribute the demand evenly. This is applicable to satisfying the different parking demands of the car drivers by the parking lots. Hence, there is a need for a decision based allocation module that takes care of even distribution. In the work proposed, we use a Cellular
Automaton model for this purpose.
Fig. 4. Sequence of the simulation for parking request A Cellular Automaton (CA) (Plural: Cellular Automata) is
a decision making module that work on a group of cells [
Using a CA, each cell’s next state is decided by it’s current
transfer is to assess the overtime parking revenue of each state and the nearby cells’ current state [
In this work, by lattice we refer to an ordered group of
individual cells [
F. Web Service and Shared Parking model Figure 7 illustrates the process of how the mapping illustrated
The cloud-based web service incorporates both the base and before is realised in the work proposed. The allocation module shared parking model. Application interfaces act as the access ensures that for a reservation, the closest parking lot that was to the web service, since it differentiates and redirects the not reserved in the threshold time (tth) duration is used. The request from RSU. The evaluation module receives the request calculation of the previous state of the parking lots starts by from the application interface and proceeds to gather and getting the state of the parking lots nearest to the destination. elaborate the data and then finally responds to the requester Obtaining parking lots nearby is taken care of by distance with the appropriate response. A database is also a part of the manipulation module. Then the field last reserved time helps web service. The database consists of details about commercial to calculate the previous state. The next step is to filter the parking lots on the basis of the threshold time. The step calculation of previous state ensures this. Rule 51 transforms the cells of the lattice into the next state in the transformation step. The vehicle demanding parking is allocated the parking lot enabled by the allocation module. In the case of multiple enabled parking lots in the lattice, the first and foremost entry in the lattice satisfies the demand generated by a vehicle.
Wolfram’s Rule 51 helps to distribute traffic evenly. A single CA represents a lattice of parking lots. 1) Emission logs help to assess vehicular emission output
of all the vehicles.
help to extract the following data:
Reservation data (to estimate utilisation) Revenue and Overtime Estimates (to estimate overall revenue of each parking lot).
We evaluate the model and its implementation on the data of the city of Dublin. The evaluation is run along five dimensions: (A) Realisablity of a simulation-testbed, (B) Utility of Vehicle to Infrastructure communication in terms of emissions, (C) Correlation between emissions and parking ease, (D) Economic benefits for the parkign owners, and (E) Effects of the number of available parking lots.
To arrive at a stable infrastructure required for the proposed
work, the performance and the applicability of various V2X
frameworks have been analysed. According to the College
of Computing, Georgia Institute of Technology and School
of Civil and Environmental Engineering, Georgia Institute
of Technology, a simulation-based testbed can be defined as
a testbed being capable of simulating different components
required for the experiment, operating the components
simultaneously as in real-world, and observing the proposed
system design changes through logs [
We consider a scenario consisting of 4 parking lots each with 2 spots as capacity and 76 vehicles, 20 of which are actively look for parking. The metrics that are used to illustrate the benefits of the proposed model are ”Mean Vehicular CO2 Emission” and ”Reroutes.” The base scenario is realized in SUMO. The scenario reflects the realistic behaviour of a driver. When the free spots are not available in a parking lot, a driver looks for another parking lot in the vicinity. This information comes to the knowledge of the driver when the vehicle is near the parking lot. Searching for another parking lot is interpreted as a reroute and this, in turn, leads to more CO2 emissions by each of the vehicles. In scenario with a communication strategy i.e. proposed model, vehicles receive information about services offered through V2X messages. The vehicle’s On-Board Unit (OBU) processes the received V2X message and instructs the navigation module of the vehicle to go to the parking lot with free spots. In both the scenarios, vehicular emissions and reroutes are extracted using the inbuilt logging features of SUMO and MOSAIC. These values are obtained after running the simulation.
The results show that a communication system makes the drivers aware of the services offered and redirects the vehicle to a corresponding parking lot with lesser vehicular CO2 emissions i.e. substantially reduced reroutes.
The results are presented in Table I and Figure 9 and they highlight that the rate of decrease in mean vehicular emission when equipping the scenario with a communication strategy is 46.75%. The number of reroutes in the proposed model scenario is 0 as the services offered are communicated to the vehicle before it reaches the destination whereas, in the case of the base model scenario, it is 16.The value 16 is obtained as a result of running the base simulation using SUMO. Reroute is a standard metric related to parking. The value increases with the number of times vehicles are getting redirected in search of a parking spot., which is part of SUMO simulations.
The vehicles moving in the simulation might have abnormal
routes which eventually change the emission values. Hence,
in order to remove this abnormality, we run the simulations
several times to consolidate the result. The hours during which
the simulation is run is between 7:00 to 18:00. The reason
behind this decision is that the demand data obtained from
[10] gives the vehicle volume within this time period. The
simulation scenario is run for each hour and the logs are
deserialized. The process is repeated for the base model and
shared model scenarios for each hour of the day separately
and the results are consolidated for further analyses. The
infrastructure for the simulation of the base model scenario
has a V2X framework simulation scenario and Python-Bottle
based Web Service [
To estimate emissions, we resort to the model proposed in
[
1) Relationship between CO2 emission and distance travelled: to establish the relationship between CO2 emission value and distance, we perform a correlation analysis. The results, presented in Figure 10), show that there is a strong linear correlation between CO2 emission values and distance i.e. with the increase in emission values, distance increases.
2) Linear regression model: The data coming from this setup is amenable to fit a linear regression model. This is useful to estimate distances for the emission data coming from the simulations.
setups, one not including private parking and one including them, are compared on the basis of CO2 emissions (milligrams) and distance travelled (metres) by vehicles in the simulation. At each time of the day, the parking demand changes. The results of the comparison are presented by Figure 11 and Table II. Figure 11 represents the comparison analysis between base model (without shared private parking lots) and shared model (with shared private parking lots) in terms of vehicular emissions and distance travelled by vehicles in the simulation. Three sets of plots are present in Figure 11. From top to bottom, each set is plotted for base model and shared model at different hours of a day. For each hour of a day, mean vehicular emission and mean distance travelled by vehicles in the simulation are plotted. Each blue bar and blue stem represents vehicular emission and distance travelled by vehicles respectively for the base model. Each orange bar and orange stem represents vehicular emission and distance travelled by vehicles respectively for the shared model.
Table II is constructed by considering vehicular emissions in all hours of a day. In Table II, the mean value of vehicular emissions of all the vehicles in base model and shared model are compared. Base model and shared model are compared and evaluated by considering the rate of decrease in emissions in Table II. The base parking model remains the same as part of the comparison for all the ratios. But based on the demand at a particular hour of a day, the output emissions and distance values change for the shared parking model due to the varying number of shared parking lots. The results show that as the number of shared private parking lots increases, the emission value decreases. The rate of decrease in vehicular emissions (when compared to the base parking model) is 2% linearly with an increase in shared parking lots of 4. Since distance travelled is linearly dependent on emission values, there is a decrease in distance travelled by the vehicles in the shared private parking model with respect to that in the base parking model.
The utilisation rate of a parking lot is the sum of all the reservation time of a parking lot over the unit of time. The expression for utilisation rate is given by Equation 1. The records from the database are interpreted for estimating the utilisation rate of each parking lot. The reservation entry is also associated with over time estimates. If the vehicle is parked over time, then there is an entry as over time estimate associated with the reservation. As per the mathematical model, reservation of a spot is done only when all the conditions are satisfied, the most important of which is the preferred distance to travel by foot for a car driver. The distance by foot from the parking lot must be lower than the maximum distance preferred by the car driver to walk. Apart from the walking distance condition, we also consider free spots availability and plan to consider additional preferences such as disability preferences and presence of charging spots. 1) Description of comparison analysis plots: In Figure 12 and Figure 13, utilisation and revenue generated are compared between the scenarios with base model (without shared parking lots) and shared model (with shared parking lots) in Figure 12 and Figure 13. Due to varying number of shared private parking lots, there are three sets of comparisons between same base model scenario but different shared model scenarios (i.e. from top to bottom in each comparison set, shared model with 24, 28 and 32 shared private parking lots).
2) Comparison analysis between commercial and shared parking lots in terms of utilisation rate and revenue: Figure 12 compares the the base model and the shared model in terms of utilisation rate. In Figure 12, each block has a set of line plots with utilisation rate in y-axis and time of day in x-axis. The set of plots include utilisation of different commercial parking lots along with combined utilisation of many shared private parking lots. ILAC parking lot is a commercial parking lot in North-Western part of Dublin. From Figure 12, ILAC parking lot (plotted in green in each block of Figure 12) is utilized more than that of the other commercial and shared parking lots in the proximity. This is due to its location being in a strategic point that makes it more preferred than any other.
The allocation module (equipped with CA) is intended to make sure that the demand is evenly distributed among the parking lots in the shared parking model (including the commercial parking lots). But in the process, due to the strategic location of some parking lots and the vehicle driver’s end destination, the strategically located parking lots are preferred more than others. This is evident from the maximised utilisation of ILAC parking lot. We analyse how shared parking lots are utilized when compared to commercial parking lots. With the increase in the number of shared private parking lots, the demand satisfied by the shared parking lots increases. This in turn decreases the utilisation of the commercial parking lots. The presence of shared parking lots contributes to a considerable drop in the utilisation of commercial parking lots, especially ILAC. This is evident from Figure 12.
In Figure 13, each block has a set of line plots with revenue in y-axis and time of day in x-axis. The set of plots include revenue generated by different commercial parking lots along with combined revenue generated by many shared private parking lots both in base model and shared model. When considering collectively, the overall revenue generated by the parking lots is proportional to the utilisation rate. As the utilisation rate increases, revenue also increases. But the randomized overtime parking behaviour of the vehicles helps the parking lots to earn more.
It is also important to note that the utilisation rate and revenue of shared parking lots are constant for all graphs of Figure 12. This is because we denote the changes in the utilisation rate of shared parking lots collectively and not individually. The constant number of shared parking lots each with 2 spots satisfy the maximum demand possible. But when there are no vacant spots within the range of the driver’s preference, the rerouting occurs. This rerouting also contributes to maintaining the constant utilisation of parking lots.
take the perspective of the single parking lot. For this purpose, 3 shared parking lots are considered. The pattern by which these parking lots are utilised varies for each simulation run, thus generating many possible configurations. In Figure 14 we show few patterns of the utilization of the three parking lots.
From the plots in Figure 14, the important inference is that each of the parking lots is guaranteed to get utilized at least twice (Parking Lot 3 in (a) and Parking Lot 1 in (c)). This is guaranteed by the allocation module using Rule 51. In the case of other parking lots, at some time of the day, the utilisation exceeds 2 and ranges between 2 to 3. This is due to their location and also to the destination of the vehicles at that time. This contributes evidence to the fact that shared private parking lots are utilized evenly. Also, the presence of such shared parking lots decreases the dependency on commercial parking lots as visible in Figures 12 and 14).
Lastly, we evaluate the effects of the ratio between shared parking lots and commercial parking lots available. The results are from the driver’s and the parking lot owner’s perspective. This collection helps us to correlate how the variation of the ratio could benefit the shared private parking model. From the simulation one can identify the optimal number of parking lots needed for the benefit of the stakeholders. Also, it further helps to understand how to extend the model from a part of a city to a full city-based scenario such as for the MoST. The parameter Vehicular Emission denotes the effects on the vehicles to changes in the number of shared parking lots from a driver’s perspective. The parameters Redirects and Supply to demand denote how the model with a certain number of shared private parking lots react to the demand (number of vehicles) at a particular hour. Figure 15 shows the results of correlation analysis in the form of a heatmap.The gradient of correlation of different parameters with each other is represented by heatmap. The gradient is represented by numerical variation between -1 and 1 and also, in the form of coloring scheme from dark green to dark red in Figure 15.
Figure 15 shows that the ratio and the mean vehicular emission of all the vehicles in the simulation are moderately correlated with a correlation of -0.32. This correlation can be best described as with the increase in the number of shared parking lots, there is a moderate decrease in vehicular CO2 emissions. From the driver’s perspective, this means that the vehicular emission due to additional cruise time to search for parking spot reduces considerably.
2) Relationship between ratio and supply to demand: the term supply to demand denotes the number of reservations made by the vehicles in the shared parking lots instead of the commercial parking lots. There is a strong positive correlation of 0.82 that shows that with the increase in the number of shared parking lots, more vehicles utilize the shared parking lots rather than resorting to the commercial parking lots.
3) Relationship between ratio and redirects: The term redirects indicates the number of vehicles that are not served by the shared parking model. The correlation heat map shows that the correlation coefficient is -0.5. The conclusion is that as the number of shared parking lots increases, the number of vehicles not served decreases. In the present simulation, varying the number of shared parking lots, the best-suited value is 32.
Smart cities try to optimize the way in which the shared infrastructure is used by its citizens, opening to the possibility of people actively participating resource utilization. The work proposed here provides a fully worked solution on how to address the problem of parking scarcity. Additional benefits of sharing private parking spots in terms of environmental and economic benefits are also identified. We proposed a setup of a simulation-based testbed to realize and analyze the shared private parking model, required number of framework researches and trial runs. The design, implementation and evaluation via simulations show the high potential benefits of the smart shared parking model in terms of CO2 emission reductions, decrease of driven distance, and economic compensation for private parking owners. Finally, an important outcome of the work is that the results demonstrate how the change in the number of shared parking lots affects the model itself and the participants of the shared parking model.
Limitations of the present study include the following ones. The scalability of the setup ensures that the applicability to different demand data varying by each hour. But the simulation has to be re-initiated for each hour. This is due to the fact that the retention of the number of free spots (of each parking lot) after respective vehicles resumed is not certain. This uncertainty is due to possible communication failures of the resume signal of the vehicles.
The parking and overtime parking behaviour of each vehicle is modelled in the simulation as random entry. The random entry is chosen by each vehicle from the set of defined entries (taken care of by the application deployed in each vehicle involved in the simulation). For example, for vehicle A, the parking duration and the resuming duration are chosen randomly from the defined set as 30 minutes and 35 minutes, respectively. Though the duration pair is a fixed one from the defined set, the method of choosing a pair from the set is random. There is a need to model the driver’s behaviour to park overtime and incur additional charges due to overtime parking which has not been addressed in the work proposed.
Apart from modelling the parking behaviour externally, the vehicle resume activity needs to be triggered externally to reflect the externally modelled behaviour. There are few standard metrics to analyse a shared parking model. In the work proposed, the beneficial perspectives of the model of a vehicle and a parking lot are quantified by vehicular emission, utilisation rate and revenue. But, from a smart city’s perspective, some additional standard metrics could be added. This will pave the way for more standard ways of evaluating a shared parking model (for example quantifying traffic congestion created due to parking searches).
Finally, the messages sent between various components in the simulation needs a standard. There are accepted standards of messages based on priority. In the work proposed, the prioritisation of the parking request and response messages is not addressed. Hence, an evaluation of how other messages related to co-operative mobility get along with the messages related to service requests and responses is in order.
The work proposed has a microscopic area as the simulation-based testbed. One can extend it to a large city as the testbed for assessing the benefits of the smart shared private parking model. The work proposed can be extended to Hardware In Loop (HIL) simulation or to a digital testbed. This helps to address the realistic uncertainties that go unaddressed due to the software simulation (e.g., hardware or firmware issues of components involved in V2X communication). The shared parking model is present as a single, centralized web service. This setup could be made as a distributed architecture with proper synchronisation of the distributed data pushing the logic to the periphery of the network in an Edge Computing fashion. Apart from considering the preferred walking distance of the car driver, one could also consider adding other forms of travel options, such as a bike, an electric scooter, or public transportation. This could further reduce the congestion in the considered area. Accordingly, there will be some liabilities for the car drivers as they use the property of the owner. The mathematical models used in the work proposed could be extended in order to evaluate the effects of the change in terms of metrics discussed in the work or by adding additional metrics. [10]