A Context-Adaptive Algorithm for Truck Stop Detection using Directional Change in Global Positioning System Trajectories

Reliable detection of truck stop events from anonymous global positioning system (GPS) trajectory data is key to inferring freight movement patterns, which inform a range of applications from origin–destination mapping to infrastructure and logistics planning. Existing stop detection methods commonly rely on external point-of-interest (POI) or network data or use fixed spatial thresholds. Use of fixed spatial thresholds can lead to merged (undercounting) or split (overcounting) stop events when applied across heterogeneous spatial settings. This study proposes an adaptive stop detection algorithm that identifies truck stop events using only GPS trajectory data, without requiring supplemental transportation network-, land-use-, or POI datasets. The method integrates time-based candidate stop identification with a context-sensitive spatial aggregation mechanism in which the stop aggregation radius is continuously adjusted according to the frequency of significant directional changes observed prior to each candidate stop. A labeled dataset of 200 trajectories (570 stops) was used for calibration and testing, with an 80/20 split. Sensitivity analysis identified a 5–10 min pre-stop window, a 25-degree turn-angle threshold, and a nonlinear shape parameter of 0.2 as the most effective settings. On the independent test set (40 trajectories, 128 stops), a fixed aggregation radius of 1,100 m yielded 7% error, while the proposed method reduced error to 4.6% (∼34% improvement). These findings demonstrated that dynamic adjustment of aggregation parameters significantly improved robustness and accuracy, providing a scalable and reliable framework for preprocessing GPS trajectory data in freight analytics.