When measuring how drivers overtake cyclists, one of the underlying problems is extracting the overtaking event from a time series of lateral distance readings. This note aims to describe a simple approach that seems effective in applications like ours. It consists of carefully transforming our problem into a network problem, then leveraging a community detection algorithm to extract subsequence candidates. Lastly, we choose the anomalous subsequence from the set of returned subsequences. To the best of our knowledge, this approach to anomaly detection does not appear in the literature even though it is intuitive, offers a fair amount of control, and is not computationally expensive. Our goal is to present the crux of the method with clarity and identify where more effort could improve it. We demonstrate our approach with modularity-based community detection and point out a shared nature of our approach with density-based cluster detection methods.