I had originally remembered posting a news link (quite a while ago) about the F.B.I. attempting to compile a database like that. But, now I can’t seem to find much in the way of reference past 2008. They probably ran into all the problems you just described. (Or it works so well they don’t want to tell anyone.)
How do you start such a database even if the analysis software did work?

Also, it seems like the more data you have in the way of “gaits”, the higher the probability for false positives.