基于信息熵的非机动车超越轨迹分段方法

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

通过自行车轨迹识别超越行为是评价非机动车交通服务水平的重要工作之一。针对基于阈值分段方法中需对不同轨迹确定不同的阈值问题,引入信息熵对非机动车超越轨迹进行分段。根据实测视频提取了780条非机动车超越轨迹数据,包括了在视频中可能存在的11种超越轨迹情形,并通过对超越过程中各阶段的特征参数分析,最终选取横向加速度、横向偏移距离、偏移角度作为基于信息熵分段的特征参数,通过引入信息熵理论,提出基于信息熵的非机动车超越轨迹分段方法和分段判断条件。根据信息熵理论中,分段后的2段子轨迹中的特征参数概率密度相较分割前更接近时熵增的定律,同时考虑非机动车超越轨迹的特征参数特征,提出适用于非机动车超越轨迹的信息熵分段标准。以实测路段非机动车超越轨迹数据为实验样本,将基于各特征参数的信息熵分段结果与基于时间、速度阈值的分段结果分别带入K最近邻(K-nearest neighbor,KNN)分类算法中进行超越轨迹识别,并利用轨迹覆盖度指标评价不同分段方法的超越轨迹分段效果。实验结果表明:基于信息熵超越轨迹分段方法的超越轨迹覆盖度平均为83.0%,优于基于阈值分段方法的轨迹覆盖度平均值79.7%,且基于横向加速度信息熵分段法的平均轨迹覆盖度为85.1%,分段效果相较于其他特征参数信息熵分段方法效果最优。

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 var-ious trajectories,this paper introduces information entropy theory to segment overtaking trajectories of non-motor-ized vehicle.Using video data,780 non-motor vehicle overtaking trajectories are extracted,and 11 potential overtak-ing 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 informa-tion 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 pa-rameters in two sub-trajectories after segmentation is closer than before segmentation.Besides,considering the fea-tures 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 ex-perimental 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 cov-erage 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 trajecto-ries,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.