Limited Sensing and Deep Data Mining: A New Exploration of Developing City-Wide Parking Guidance Systems

With the increasing automobiles in China, the parking difficulty has gradually spread over the city. Developing city-wide parking guidance systems (CPGS) has become urgent to many local governments of China nowadays. However, there haven't been any effective systems available yet, because the economic investments on sensors and the time costs of negotiation with proprietors of parking lots are unaffordable from city-wide perspective. The lack of parking data is the most critical problem in development of CPGS. In this paper, we propose a new lightweight technical solution to implement CPGS by strengthening data mining and utilization under the current conditions. We start with mining the public information of parking lots, to quantify the service capability and pick out some of the most important as the samples. Next, we build a RGAN to generate data for those parking lots without data, patching them based on the samples with the similar surroundings. The technology of data generation will help us rebuild the city-wide parking data at low costs, both time and money. Finally, we design a model of recommendation based on the real-time driving data to implement intelligent parking guidance, featuring active recommendation and automatic adjustment with vehicle moving. The solution makes the best use of the existing data while reducing the reliance on sensors significantly. And we develop an experimental system in Shenzhen, achieving the rather satisfying guidance effect in practice. Hence, it's a novel and effective exploration of CPGS, breaking through the dilemma of heavy dependence on sensors.