Depending on coverage, this might not be enough. People with masks can be tracked as well. E.g.: At some point, persons A, B and C enter an area without surveillance; Person B and C come out, A remains unseen but new person X comes out, wearing a mask. Even if they all don't leave during the period when X is out destroying government property, the problem is reduced from finding the needle in a haystack to identifying a ~1.75m lean person with a brisk walk, blue Sneakers and specific political views out of a relatively small set of individuals.
And obviously dance all the way to fool gait recognition.