I'm a senior software developer with over 6 years of experience in computer vision field. I'm also a PhD student, my areas of expertise are human detection and tracking algorithms, object recognition and OCR systems using neural networks. In this project I would suggest combining two mechanisms: first is a simple face detector, for exmaple using Haar wavelets and the second one would be a Kalman or particle filter, which will estimate next coordinates of the head (basing on previous ones). Moreover, I think it would be better to use EmguCV, which is a complete, stable .NET wrapper for OpenCV. If you have any questions, please feel free to contact me.