Here is another "simple" explanation:

https://rweba.livejournal.com/446326.html

It's definitely WAAAY oversimplified, but it gives the first idea:

The "filter" just cleverly averages the noisy real data at time T with the Kalman estimate from time (T-1).