Dynamic time warping (DTW) is a dynamic programming approach used to align multi-view time series. The similarity between two sequences are measured, and used to find an optimal match between them via time-warping (inserting frames).
This approach requires timesteps in the two sequences to be comparable, and requires a similarity measure between sequences. Because DTW requires a pre-defined similarity measure, it has been extended to Canonical Correlation Analysis which does this computation in a coordinated space for Multi-modal Machine Learning. This allows for both aligning and mapping between different modalities jointly in ean unsupervised manner.