Multi-view Low Rank Representation for Multi-Source Traffic Data　Completion

摘要

Intelligent Transportation System (ITS) has been widely applied in major cities to relieve congestion and decrease accidents.However, the hardware failure of detectors or transformation failure of data cause data loss, which seriously decreases theperformance of ITS. How to ensure the completeness of observed traffic data becomes is a current key problem. Recently,the low rank constraint which can exploit the global relation hidden in data has been successfully used in matrix completion,such as the classic robust principal component analysis (RPCA) and its variants. The spatio-temporal correlation amongtraffic data make traffic data contain low rank property; therefore, we naturally apply the low rank constraint on trafficdata completion. In addition, most traffic detectors installed on the road can collect various types of traffic data, so-calledmulti-source traffic data. Due to describing the same traffic condition, these various type of traffic data usually have similarintrinsic structure. Therefore, we consider fuse these various type of traffic data to complete the missing data. In thispaper, we propose multi-view low-rank representation model for multi-source data completion and provide an efficientoptimization algorithm. To variety the performance of the proposed method, some traditional traffic data completionmethodsare compared with our method on a highway microwave dataset. The experimental results show that our proposed methodis obviously superior to other state-of-the-art traffic data completion methods.