基于V-支持向量分类机的交通事件检测方法研究

Research on Methods of Traffic Incidents Detection Based on V -Support Vector Classification

基于支持向量机在解决分类问题的优势，本文提出基于V-支持向量分类机的交通事件检测方法。首先把交通事件是否发生看成是一个特殊的分类问题，选取V-支持向量分类机和核函数，根据以往的交通事件是否发生的检测数据，即分别在发生交通事件和不发生交通事件两种情况下的上下游车道占有率，计算出其当前时段的上下游车道占有率的绝对差、相对差，以及下游前两时段与当前时段车道占有率的相对差，以此作为V-支持向量分类机的输入，对其进行训练，然后输入现阶段检测到的相应车道占有率统计结果，利用训练完成的V-支持向量分类机来判别是否发生交通事件。最后，本文以微观交通模拟的数据验证模型的效果。

Because support vector machine excels at solving classification problems, the paper proposes methods of traffic incidents detection based on V-Support Vector Classification. First, whether traffic incidents occur or not is considered as a special classification problem. When the type of V -Support Vector Classification and its kernel function are selected, according to the collected lane occupancy-rates under normal condition and under the condition that traffic incidents occur, that is, the lane occupancy-rates of upstream and downstream under normal condition and under the condition that traffic incidents occur, the difference and the relative difference between data of upstream and that of downstream in the same interval or the relative difference between downstream data in the previous interval and that of the next interval can be calculated, which are set as the input of V-Support Vector Classification. Then after inputting current corresponding statistical results of lane occupancy-rates, traffic incidents can be detected by means of the trained V-Support Vector Classification. Finally, the paper will use data based on microcosmic traffic simulation to test the efficiency of the proposed methods.