Dear Pan,
There are so many easy and complex ways. The best and the simplest is based on Entropy.
You calculate the Entropy of each frame and then find the shots. Like whenever the difference between two consecutive frames is more then some Threshold then mark it as shot. So the median frame can be good representation or take the frames having maximum Entropy.
for key frame you can refer to many papers. The method I explained you can be read in detail in following paper.
Video segmentation into scenes using entropy and SURF
