Contents I. Introduction
Traffic congestion has become a universal phenomenon in metropolitan cities. It is estimated that commuters searching for parking during peak hours contribute around 30% of the traffic flow in a city [1]. Hence, it is not surprising that one important aspect of traffic management is street parking. Driving aimlessly around city blocks looking for a vacant car parking space not only leads to increased traffic congestion, which translates to waste of fuel and increase of pollution, but also to significant reduction in productivity [2]. Thus, in order to deploy the vision of smart cities, the challenging problem of parking must be alleviated by establishing a highly functional and efficient street parking detection system, which can warn drivers of available vacant parking spots in the street ahead [3]–[4]. Initially, sensor-based systems that rely on ultrasonic [5] or wireless-magnetic [6] based sensors installed on each parking space or sensor-fusion based networks [7] were used to identify vacant parking spaces in parking lots. However, these methods cannot be implemented for the on-street parking task. In [8], the authors introduce a scheme that uses data obtained from video surveillance cameras in an urban environment for the detection of parking spots using Support Vector Machines (SVM) and k-nearest neighbors algorithms for identifying parking. Although this model produces a good level of accuracy, the method uses an aerial view of the block rather than a direct street view, making it very impractical. In summary, to the best of our knowledge, existing parking detection methods do not address the challenge of identifying available street parking spaces.
