Two-dimensional dense seismic ambient noise array techniques have been widely used to image and monitor subsurface structure characterization in complex urban environments. It does not have limitations in the layout under the limitation of urban space, which is more suitable for 3D S-velocity imaging. In traditional ambient seismic noise tomography, the narrowband filtering (NBF) method has many possible dispersion branches. Aliases would appear in the dispersive image, and the dispersion curve inversion also depends on the initial model. To obtain high-accuracy 3D S-velocity imaging in urban seismology, we developed a robust workflow of data processing and S-velocity tomography for 2D dense ambient noise arrays. Firstly, differing from the NBF method, we adopt the continuous wavelet transform (CWT) as an alternative method to measure the phase velocity from the interstation noise cross-correlation function (NCF) without 2π ambiguity. Then, we proposed the sequential dispersion curve inversion (DCI) strategy, which combines the Dix linear inversion and preconditioned fast descent (PFD) method to invert the S-velocity structure without prior information. Finally, the 3D S-velocity model is generated by the 3D spatial interpolation. The proposed workflow is applied to the 2D dense ambient seismic array dataset in Changchun City. The quality evaluation methods include residual iteration error, horizontal-to-vertical spectral ratio (HVSR) map, and electrical resistivity tomography (ERT). All tests indicate that the developed workflow provides a reliable 3D S-velocity model, which offers a reference for urban subsurface space exploration.