Low Rank Aligning Two Point Sets for Unsupervised Correcting GPS points of Parking-slots

Abstract

Parking management is an important part of urban management. Accurate Global Positioning System (GPS) points of parking-slots are the core of various applications ranging from looking for vacant parking-slots to designing the parking policies. Affected by high-rise buildings and various signals, either blocked or affected GPS points tend to mismatch the actual locations. Therefore, due to the large number of parking-slots, it is not a trivial problem to correcting raw GPS points in a unsupervised approach. In this paper, discovering that the parking-slots always most parallel with the roads in a city, we proposes an unsupervised low rank method to effectively correct GPS points of the parking-slots by rotating-translating-aligning operations. The proposed method theoretically handles any errors of GPS points; besides, the proposed unconventional alignment method is simple, yet effective and efficient for large-scale points for a city. Extensive experiments demonstrate the superiority of the proposed method.