Compressive Sensing Approach to Urban Traffic Sensing

Traffic sensing is crucial to a number of tasks such as traffic management and city road network engineering. We build a traffic sensing system with probe vehicles for metropolitan scale traffic sensing. Each probe vehicle senses its instant speed and position periodically and sensory data of probe vehicles can be aggregated for traffic sensing. However, there is a critical issue that the sensory data contain spatiotemporal vacancies with no reports. This is a result of the naturally uneven distribution of probe vehicles in both spatial and temporal dimensions since they move at their own wills. This paper proposes a new approach based on compressive sensing to large-scale traffic sensing in urban areas. We mine the extensive real trace datasets of taxies in an urban environment with principal component analysis and reveal the existence of hidden structures with sensory traffic data that underpins the compressive sensing approach. By exploiting the hidden structures, an efficient algorithm is proposed for finding the best estimate traffic condition matrix by minimizing the rank of the estimate matrix. With extensive trace-driven experiments, we demonstrate that the proposed algorithm outperforms a number of alternative algorithms. Surprisingly, we show that our algorithm can achieve an estimation error of as low as 20% even when more than 80% of sensory data are not present.