An Improved Low Rank Matrix Completion method for Traffic Data

Intelligence Transportation System (ITS) plays a more and more important role in the daily life, such as transportation route planning, vehicle scheduling, prediction and so on. The speed data of the roads are the major data. However, the road speed data always contain partial missing data, which affect the performance of Intelligent Transport System. These missing traffic data are usually complemented by the current speed values of neighbor roads or the historical speed values of the road, but the completion results are not satisfied. Since traffic data has spatial-temporal correlation, some researches employ low rank matrix completion method to recover the traffic missing data and achieve wonderful experimental results. In actually, analyzing the traffic data, we can even know the exact rank of the traffic data, which is ignored by traditional nuclear norm based traffic data completion methods. We can firstly replace the nuclear norm with partial sum minimization of singular values constraint in conventional low matrix completion methods for traffic data completion. Then, we further to find the strong spatio-temporal correlation exist in traffic data; as a result, we introduce order constraint into our proposed method. A proper solution for the proposed methods is given. The proposed method is tested on two real road speed database, which demonstrate our proposed method is superior than the traditional traffic data completion methods.