Computational analysis of the rhythmic/metrical structure of music from recorded audio is a hot research topic in music information retrieval. Recent research has explored the explicit modeling of characteristic rhythmic patterns as a way to improve upon existing beat-tracking algorithms, which typically fail on dealing with syncopated or polyrhythmic music. This work takes the Uruguayan Candombe drumming (an afro-rooted rhythm from Latin America) as a case study. After analyzing the aspects that make this music genre troublesome for usual algorithmic approaches and describing its basic rhythmic patterns, the paper proposes a supervised scheme for rhythmic pattern tracking that aims at finding the metric structure from a Candombe recording, including beat and downbeat phases. Then it evaluates and compares the performance of the method with those of general-purpose beat-tracking algorithms through a set of experiments involving a database of annotated recordings totaling over two hours of audio. The results of this work reinforce the advantages of tracking rhythmic patterns (possibly learned from annotated music) when it comes to automatically following complex rhythms. A software implementation of the proposal as well as the annotated database utilized are available to the research community with the publication of this paper.