Explore the lane change trajectory pattern driven by full-time domain trajectory data

Due to the increasing density and complexity of the highway network, a deep understanding of the characteristics of lane changing (LC) behavior is crucial for road refinement design. The emergence of full time domain trajectory big data provides unprecedented opportunities for in-depth research on highway safety geometry design. This article proposes a method framework for extracting LC trajectory patterns to explore the combination trajectory patterns during the LC process. This article achieves subdivision in driving mode detection by using the adaptive pruned exact linear time (APELT) algorithm to detect change points, taking into account the short sequence features of LC. In order to achieve the classification of segmented fragments, we report a clustering technique based on similarity matching (SM), which can effectively avoid the problem of excessive distortion in similarity measurement. The results indicate that APELT technology has certain advantages in F₁ score and accuracy of LC pattern recognition, which is more in line with reality. The kappa score based on similarity matching is greater than 0.8, indicating high accuracy of pattern recognition. This study provides a novel data mining method for a comprehensive understanding of lane changing behavior under full time domain big data, which will provide reference for road design in complex scenes.