Segmentation of Overtaking Trajectories for Non-motor Vehicles Based on Information Entropy

Identifying overtaking behavior through bicycle trajectories is essential in evaluating the service level of non-motor vehicle transportation. Threshold-based segmentation methods require setting different thresholds for various trajectories, this paper introduces information entropy theory to segment overtaking trajectories of non-motorized vehicle. Using video data, 780 non-motor vehicle overtaking trajectories are extracted, and 11 potential overtaking scenarios are covered. By analyzing the characteristic parameters of each stage of the overtaking process, lateral acceleration, lateral offset distance, and offset angle are identified as the characteristic parameters based on information entropy segmentation. A method for segmenting overtaking trajectory of non-motor vehicles is developed using information entropy theory, and the segmentation judgment criteria is proposed based on this theory. According to the information entropy theory, the law of entropy increase indicates that the probability density of characteristic parameters in two sub-trajectories after segmentation is closer than before segmentation. Besides, considering the features of characteristic parameters of non-motorized vehicle overtaking trajectories, the information entropy segmen-tation standard is proposed for non-motorized vehicle overtaking trajectories. Taking the real trajectory data as experimental samples, trajectory segmentation is carried out using the information entropy segmentation method, and baseline methods with time and speed threshold, respectively. K-nearest neighbor (KNN) classification is adopted for recognizing overtaking trajectories based on the results of trajectory segmentation. Moreover, the trajectory coverage index is used to evaluate the effectiveness of different segmentation methods. The experimental results show that the information entropy based segmentation method has an average coverage of 83.0% for overtaking trajectories, compared to a coverage of 79.7% for the threshold based segmentation method. The information entropy based trajectory segmentation method outperforms the threshold based trajectory segmentation method. Furthermore, the average coverage of lateral acceleration of information entropy based segmentation method is 85.1%, achieving the best performance among the information entropy segmentation methods with different features.