Finding parking is a worldwide problem experienced by 92% of drivers, according to the recent 2023 Global Driver Survey. As a result, the accuracy of parking information is crucial for drivers, with the value of connected car services continuing to rise each year. Dynamic Data plays a crucial role in providing ever more accurate parking availability data for drivers to determine how likely they are to find suitable parking at the end of a journey. Parkopedia’s Chief Engineer, Ganesh Sivalingam, explains how it all works.
When founded, Parkopedia originally offered solely ‘static parking data’ to customers. This consists of core information that is generally unchanged and can be known ahead of time, such as opening hours, location and the total number of spaces. Parkopedia’s extensive database has taken over a million working hours to compile and is continually monitored and updated to maintain accuracy and completeness.
Even today, knowing this static data is extremely helpful for drivers looking to find somewhere to park, however, this data is unable to help drivers assess whether there will be an available space for their estimated arrival time.
The current landscape of live parking availability data reveals significant deficiencies. Merely 5% of off-street parking locations generate real-time space availability data, and this data is often incorrect. In addition, the overwhelming majority of street parking spots lack any sensors, leaving drivers without real-time occupancy information. As a result, drivers often find themselves aimlessly circling, wasting time, fuel, and increasing traffic congestion.
Even when drivers are fortunate enough to receive accurate real-time data from car parks, it is often outdated by the time it reaches them. Due to the inherent delays in data collection and transmission, by the time drivers receive parking availability information, it may be minutes old, rendering it less reliable and potentially leading to frustration and inconvenience. Additionally, real-time data fails to provide foresight into parking availability at the time of arrival, leaving drivers uncertain about their prospects and unable to plan their journeys effectively.
In addition to the practical and cost limitations of installing sensors in all parking facilities, many drivers want to know the likely availability for their arrival time, not the current availability of spaces - so they can navigate straight to the best parking location.
Artificial Intelligence (AI) and Machine Learning (ML) can be used to fill these gaps. The data mentioned can be combined with many other data sources to provide accurate predictions of parking availability at locations worldwide and at any time of day.
The key to this is understanding the base availability patterns of parking locations, achieved through ‘ground truth observations’. Parking infrastructure sensor data, as discussed above, is one form of this ground truth data, but Parkopedia also collects millions of observations with methods such as computer vision technology and our Parkopedia indoor mapping vehicle. This allows for a great understanding of the underlying availability trends of particular parking locations.
The observations collected can be associated with changes in behaviour seen in live streaming data, in order to learn what happens in abnormal circumstances - from public holidays to large-scale sporting events and even pandemic lockdowns. These are nearly infinite in variation and impossible to know ahead of time everywhere in the world, however, this is where vehicle sensor data is then utilised.
With the sharp increase in connected vehicles, numerous vehicle sensor data sources are now available. One such source is floating car data (FCD), which logs the approximate GPS position of a vehicle as it drives along (anonymised and with the driver's permission granted). This data is extremely useful for gauging the number of vehicles in a specific area and how busy it is compared to the average for that particular day and time.
Vehicle sensor data, however, is not a “silver bullet”. A motorway traffic jam, for instance, could make it look like there are a significant number of drivers seeking local parking, when in fact, the vehicles are simply passing by. Additionally, due to the privacy considerations mentioned above, this type of data is not widely available.
While predicting parking availability, the most important part of a journey is when the car pulls out of a parking space, indicating that a space has become available, and when the driver pulls up and gets out, indicating that a space has been taken. This is known as Park-In/Park-Out (PIPO) data. It is much more privacy-friendly, as all that is needed for each PIPO event is a single GPS point and a timestamp, with no need to link it to a particular vehicle, or to track over time. Parkopedia's machine learning models monitor many geographic regions at multiple levels of granularity and assess the balance of park-in and park-out events, with this ratio giving an indication of whether there will be an increase or decrease in available parking spaces in the next few hours.
The model can compare the absolute numbers of events and the ratio of vehicles parking/leaving spaces to what is normal for the day and time for this area, assessing these against the previous ground truth data. It learns the relationships between parking locations and geographical regions across the world, which enables it to understand certain patterns automatically. Consequently, based on the streaming data it can predict that there is likely to be high parking availability in urban shopping areas on Christmas Day, for example. Similarly, the technology is also automatically aware that in 2022 the UAE changed its weekend from Friday and Saturday to Saturday and Sunday, which had a knock-on effect on parking availability.
If you are wondering why we don’t just directly use PIPO data to highlight available spaces for drivers, read our previous blog post on ‘Why predictions beat real-time, every time’ for the full answer.
To address parking availability concerns, Parkopedia created its ‘Dynamic Data’ product, which reports information regarding the availability of parking locations around the world.
Parkopedia offers two types of parking availability predictions. The first is the probability of finding a single space at a parking location (with additional adjustment for larger car parks to account for the fact that finding a single space in a large car park is harder than in a small one). This indicates whether a driver will be able to successfully find an available space for their car at the location. This information is also relative to the size of the car park, where more spaces can become available in larger locations.
The second type of availability prediction is then the probability per individual space in the car park. This is not impacted by the number of spaces the car park has, and can also be thought of as an expected percentage availability of spaces at the location. This means that when you have high availability, you expect many spaces to be available, whereas with probability per location, there may only be one space.
Parkopedia has also developed ‘Parking Situation Reports’. These show aggregate parking information for a particular geographic area including both static and dynamic data and can be used with voice assistant technology to provide insight such as, “There are many parking locations near your destination, however, the majority are currently full.” It can also be used in prompts from in-car head units, on occasions when finding parking is particularly difficult, for example, triggering more advanced parking finding features, which may not be necessary if there was an abundance of available parking.
EV charging is also directly affected by parking restrictions and availability, due to the time cars need to be parked whilst charging. Data indicating the current status of charge points is often available, much more so than with standard parking spaces, due to EV chargers typically being connected devices. This means there should be greater data availability for EV-specific charging and parking spaces, giving EV drivers a better insight into their chance of finding parking and charging options. Parkopedia further supports the transition to EVs by using our dynamic data to report EV charging space availability and continue to develop new product features for the automotive industry to leverage this data further to help reduce public charging anxiety.
Various research unanimously shows that drivers worldwide highly value accurate parking information, with this being recently ranked as the most important connected car feature. Many variables affect parking availability, however, this requires a combination of both static and dynamic parking data, with these working together to gauge current availability levels and accurately predict future levels to provide the optimum experience for drivers.
Parkopedia utilises both static and dynamic data for this very reason, enabling OEMs to direct their drivers to areas with the greatest likelihood of available parking and offering features such as Parking Search Routes, which navigates drivers to their destination in the shortest possible time, while passing several locations that offer the highest chance of being able to park successfully. Parking Situation Reports also maximise this data, giving a high-level summary of predicted availability to make it easier for drivers to find parking.
Such services take the stress out of finding parking for drivers, elevating the in-car experience and offering scope for automakers to boost customer loyalty through seamless in-car connected services. With the global shift towards EVs continuing to pick up momentum and the subsequent demand for charging stations combined with a dwindling parking supply, this is only likely to accelerate the importance of connected car services covering both parking and charging in the future.
Visit our dedicated product page to learn more about Dynamic Data or to discuss further with the team.
As Chief Engineer, Ganesh is responsible for technology leadership at Parkopedia. This includes ensuring that engineering and machine learning solutions are well architected, highly scalable, properly monitored and resilient as well as improving productivity across the engineering department.