City-scale roadside electric vehicle parking and charging capacity: A deep learning augmented street-view-image data mining and analytic framework

In response to the escalating sales of electric vehicles (EVs), roadside parking and charging have been developed to facilitate EV penetration in many cities. However, its city-scale capacity is usually unknown, hindering effective planning of parking and charging infrastructures. To tackle this problem, we develop a deep learning augmented street-view-image (SVI) data mining and analytic framework, consisting of three hierarchical modules. The first module retrieves geo-locations along roads in the government authorized parking zones (APZs) and obtains SVIs that present both sides of roads centralized at these geo-locations, which is used to identify suitable roadside parking locations. The second module conducts transfer learning that determines a suitable SVI dataset with well-defined classes of interested street-view geo-objects and obtains the optimal DL model capable of refined segmentation of various types of roads. The third module identifies different urban functional zones to suggest locations suitable for roadside parking, develops a 3D space geometric projection method that estimates parking areas in each location, and unravel roadside charging capacity through geospatial statistics of existing EV charging records. As a case study using 55,724 SVIs in Singapore, the IoU of segmented avenues, paths, and sidewalks is as high as 92.51 % to 89.71 %, and we suggest 54,812 roadside parking stalls available from 6761 locations in the APZs, which can support up to 590,315 kWh/day and 5,685,923 kWh/day in the commercial zones and residential zones, respectively. Our study is significant in fundamental geospatial data construction and scaling roadside EV parking and charging in dense urban areas.