A frame is one complete still image inside a video. Play 24 of them per second and you get cinema; play 60 and you get a sports broadcast or a video game. The brain blurs them into motion because each frame appears briefly and the eye holds onto the previous one for a fraction of a second.
In a modern (progressive) video, one frame is one full picture — exactly what you'd see if you paused playback. In older interlaced TV broadcasts each "frame" was assembled from two half-frames called fields, captured at slightly different moments; this is the source of those combing artefacts you sometimes see on old VHS tapes.
For an encoder, not all frames are equal. Some frames are coded from scratch (i-frame, the "keyframes" you seek to), others are predicted from neighbours to save space (p-frame and b-frame). Their pattern — the GOP — is the single biggest reason a 90-minute movie at HD quality fits in ~3 GB instead of the ~150 GB it would take uncompressed.
