Sampling is the process of turning a continuous real-world signal into discrete values that a computer can store. Light hits a sensor; the sensor samples its brightness at a grid of millions of points; those values become a frame. Time keeps flowing; the camera samples the scene 24 or 30 or 60 times per second; those frames become a video. Every digital video is fundamentally the result of sampling reality at finite resolution in space and in time.
The fundamental constraint on sampling is the nyquist-theorem: you need to sample at least twice as fast as the finest detail you want to capture. Sample slower and the details collapse into aliasing — strange moire patterns, jagged edges, "wagon wheels spinning backwards". This is the root reason 4K cameras don't just produce "more pixels" than 1080p ones; they let you faithfully capture finer real-world details that 1080p sampling would mangle.
For a product team, sampling shows up in three places. Resolution is spatial sampling — more pixels means finer detail capturable. Frame rate is temporal sampling — more frames per second means faster motion capturable cleanly. Chroma subsampling (4:2:0, 4:2:2, 4:4:4) is under-sampling the colour part deliberately, taking advantage of the eye's lower sensitivity to colour detail to save data. All three are about deciding how aggressively to sample reality and where to spend your bandwidth budget. Get the sampling decisions right at capture and the rest of the pipeline is doing fine work; get them wrong and no amount of bitrate later will fix the resulting artefacts.
