Real-Time Parking Availability Classification on a Large-Area Scale

In urban areas around the world, drivers face the daily challenge of finding a parking space. Unfortunately, these sought-after spaces, located close to their destination, are often either impossible to find, or excessively expensive, resulting in longer search times and increased congestion in city centers. The answer to this persistent problem is an intelligent parking solution. They provide drivers with real-time access to information on parking space availability, gathered through various sensing techniques such as crowdsourcing, parking meters and sensors. Some of these systems also offer opportunistic services, such as forecasts, to adapt to unforeseen dynamic situations. Drivers’ biggest concern is find ing a spot, not knowing the exact number of available spaces or availability rate. Typically, these two parameters are estimated using regression or image processing techniques. While such solutions guarantee high predictive accuracy, their large-scale deployment is hampered by computational and data collection costs. This paper therefore proposes a new approach combining clustering and classification models to predict parking availability. Our aim is to test new methods that are relatively simple and less expensive in terms of both processing costs and amount of training data. Experimental results have proved promising, with accuracy predictions exceeding 0.84.