Deriving driver-centric travel information by mining delay patterns from single GPS trajectories

Crowd-sourcing approaches for generating accurate real time travel information for road networks is promising but still challenging. For example, travel speeds, even if derived from highly sampled GPS trajectories, have limitations in their interpretability for more sophisticated travel information such as traffic-related delays or level of service (LOS) information. The proposed algorithm in this work analyzes the flow characteristics of individual vehicles by deriving and classifying delays into LOS relevant (e.g. queuing traffic) and LOS non-relevant delay patterns (e.g. stopping at a crosswalk). In contrast to other approaches, the proposed algorithm works on single GPS trajectories collected from individual vehicles (e.g. floating car data - FCD), without the necessity to average travel speeds or travel times of multiple vehicles for reliable LOS estimation. Applied to sample GPS trajectories from test drives the algorithm reaches an overall recognition rate of 82.0% for delay classes slight delay, massive delay, single stop, and multi stops. The recognized delay patterns are capable to distinguish between LOS relevant and LOS non-relevant delays at high accuracy for subsequent delay and LOS information. The recognition of LOS non-relevant single stops reaches a rate close to 100.0%.