We describe a new method involving wavelet transforms for deriving the wind velocity associated with atmospheric turbulent layers from Generalized SCIDAR measurements. The algorithm analyses the cross-correlation of a series of scintillation patterns separated by lapses of Δt, 2Δt, 3Δt, 4Δt and 5Δt using wavelet transforms. Wavelet analysis provides the position, direction and altitude of the different turbulent layers detected in each cross-correlation. The comparison and consistency of the turbulent-layer displacements in consecutive cross-correlations allow the determination of their velocities and avoid misidentifications associated with noise and/or overlapping layers. To validate the algorithm, we have compared the velocity of turbulent layers derived on four nights with the wind vertical profile provided by balloon measurements. The software is fully automated and is able to analyse huge amounts of Generalized SCIDAR measurements.