One of our favorite things to do on a Friday evening is go to downtown Seattle and find a nice restaurant.  But this experience is always marred by my anxiety around finding a parking space.  Sure, there are many off-street locations where I could park, but they’re expensive and really add to the cost of our night out.  We usually end up circling the block looking for a spot to open up, and at the same time adding to congestion and pollution.  In fact, drivers searching for parking are one of the largest contributors to congestion in a city. INRIX is dedicated to improving driver experience and has some clever ideas for this problem.

The INRIX On-Street Parking service shows the driver which blocks are most likely to have open parking spots.  This is done by coding each side of the street with a color representing the level of occupancy of parking spots – green for low-occupancy, yellow for medium- and red for high-occupancy.  With this information displayed on a map, a driver can find a spot without blindly driving in circles.

How does INRIX do this?  If each parking spot was instrumented with a sensor the problem would be easy – we could just display a color-coded average of the number of occupied spots.  In fact some cites have attempted this solution.  In a pilot study between 2011 and 2013, San Francisco installed thousands of such sensors.  However, the sensors turned out to be less reliable than expected and the city has since ended the study.  Instead, building on INRIX’s expertise in aggregating and leveraging intelligence in diverse data sources, INRIX On-Street Parking combines data from cities, mobile payment companies, real-time parking data, connected car-sharing services, and its vast supply of real-time vehicle GPS data.

However, meter transaction data from cities and mobile payment services only tells part of the story.  By careful bookkeeping of payment times and durations, an occupancy model can be calculated.  Or course this payment information doesn’t capture those cases in which a driver leaves a spot before the time has expired.  Nor does it capture numerous exceptions that affect parking availability.  Near hospitals and city offices, many spots can be occupied by vehicles with passes permitting free-parking.  In particular in some states, parking is free for vehicles with disability permits.  And of course, some drivers may park without paying for a spot if their stay is short and they hope to avoid meter enforcement.  This problem grows later in the day as evening approaches.  This of course is the largest problem with a transaction-only model: most cities only require payment during a period from about 8am to 6pm (of course this varies from city to city – here in Seattle the paid-period extends to 8pm).  How can we calculate occupancy in the evening and overnight hours when there is no transaction data?

INRIX has a huge portfolio of real-time data supplied from our constantly expanding network of 250 million vehicles and devices.  Derived from navigation and traffic application providers who partner with INRIX, the data includes vehicle location, speed and heading.  In order to create an occupancy model when no sensor to transaction data exists, INRIX processes this data into individual anonymized trips, each with distinct starting- and ending-points.  While the inaccuracies of GPS limits our ability to be absolutely sure a single driver has stopped at a particular meter or left from a particular meter, by defining a small zone around the meter we can get a pretty good idea about arrivals and departures – a model that in our initial tests has achieved 80 percent accuracy.  The zone size has been optimized to capture as many arrivals and departures as possible while not being so large as to capture driving behavior unrelated to parking.  Also, by paying attention to interesting things like how slow the vehicle was moving before the last point, and what streets it has been on, the data can be filtered to just those cases that matter to parking occupancy.  For multi-slot zones, taking the difference between the count of arrivals and departures we can get a good estimate of how many vehicles are remaining in the zone.  Of course this data only gives a snapshot of what has happened near a station.  By recursively summing the net vehicles in the zone over time, an accurate account of occupancy can be computed.  Importantly, this model can be used both during the free-period and during the daytime paid-period, permitting both verification of the paid-period model and calibration of the free-period model and increasing confidence in the accuracy of the service.  Of course the model isn’t perfect – there are cases that can deceive the estimates.  For example, suppose an Uber driver drops a client within a meter zone.  This could be detected as a parking event, but with careful filtering on the duration that the vehicle remains in the zone, these kinds of errors can be avoided.

INRIX On-Street Parking has been developed with an eye towards global markets.  Designed with data from many cities, including Amsterdam, Cologne, San Francisco and Seattle, the product accounts for the kinds of behavior seen in busy commercial districts, sleepy residential neighborhoods and city blocks with active night life.

INRIX On-Street Parking is just the beginning of the services that INRIX can offer for drivers like me who dread parking in urban areas.  Real-time impact of events, changes in availability due to construction and time-based restrictions, and prediction of parking availability at a future arrival time are just some of the features we expect to offer as the service matures.

Guest post by:

Chris Scofield, INRIX Principal Scientist