SEER: Metropolitan-Scale Traffic Perception Based on Lossy Sensory Data

Abstract

Intelligent transportation systems have become increasingly important for the public transportation in Shanghai. In response, Shanghai Grid (SG) aims to provide abundant intelligent transportation services to improve the traffic condition. A challenging service in SG is to estimate the real-time traffic condition on surface streets. In this paper, we present an innovative approach SEER to tackle this problem. In SEER, we deploy a cost-effective system of taxi traffic sensors. These taxi sensory data are found to be noisy and very lossy in both time and space. By intensively mining the spatio-temporal correlations along with the evolution of traffic condition, SEER provides wealthy knowledge to setup statistical models for inferring traffic condition when they cannot be directly calculated. As an example, we demonstrate utilizing multichannel singular spectrum analysis (MSSA) to iteratively produce estimates of traffic condition in a metropolitan scale. The optimal window width of MSSA is determined with the basic periodicity found in traffic condition. Moreover, we minimize the number of channels required by MSSA to estimate traffic condition at any location. Given a desired estimation granularity, we optimize the MSSA parameters to minimize the estimation error.