Systems and methods for tracking persons by utilizing imagery data captured by a plurality of on-road vehicles. A large number of different persons appear in a corpus of imagery data collectively captured by a plurality of on-road vehicles. An initial and limited-accuracy model of one of the persons is used to search and detect visual occurrences of that person in the corpus of imagery data, thereby starting to track that person, in which such search is limited at first to a confined geo-temporal range, in order to limit the number of different persons over which the initial and limited-accuracy model has to search and detect that person. When the visual occurrences of that person are found, a better model can be constructed, which can now be used to expand the geo-temporal range over which yet additional visual occurrences of that person are found and used to better track that person.