In this paper, we will introduce a fusion-based system composed of real-time traffic state surveillance. This system can realize the real-time traffic state estimation with over 10,000 bidirectional road sections, all the links of the Shanghai urban road network. The system consists of three modules: SCATS data processing (MS), GPS data processing (MG), and Fusion Module (FM). In MS, traffic information collected by SCATS is converted into a kind of link-based spatio-temporal mean-speeds. Similarly, the taxis'information is also calculated for the spatio-temporal mean-speeds by means of a GIS-T map in MG. The spatio- temporal mean-speed here is defined by the mean-speed of all the vehicles running on a link during a period of time. Finally, the mean-speeds from these two sources are fused using an improved evidential fusion model.