Didn't use it myself but from what I see on internet it is just a simple database of videos. So what you was is to take a video sample and compare features with a suspect.

If for tracking (not sure if I'm not mistaking but) I would probably same location and time of video, mark it on map (maybe vector or density plot would work) and narrow the search. Performing deeper analysis is area in which suspect was detected might tell his habits or even hiding place.

As you don't want to extract features again and again, so you might need to save some features which you will use for comparing with videos. I might compare images of similar time. E. g., if you have image from 2018 and you already detected person in video of 2022, you might want to search based on that 2022 data to detect him again in 2021-23 time interval.

Again, just some ideas and some masters in this field might offer something better but its a starting point I guess.