Dynamic time warping vs windowed cross-correlation vs wavelet coherence?

Hello,

I have video data with people talking with each other and I'm planning to analyse patterns of their head movements and facial expressions. I'm not very advanced in terms of statistics, so could you please recommend me one of these 3 approaches? What would be best: dynamic time warping, windowed cross-correlation and wavelet coherence? I guess windowed cross-correlation wouldn't work here, because of the stationarity assumption?

Thank you!

Paula