基于多SVM分类器融合的高速公路异常事件检测方法

提出了一种利用多SVM分类器对高速公路中的复杂交通信息进行有效融合的异常事件检测方法.首先,将初始训练集划分为互不重叠的子集,为每个子集训练分类器.给定一个输入向量,利用分类器求得其所属的类别标签,并计算出该向量对特定簇的隶属度.其次,利用概率方法将多SVM分类器分类结果进行融合,得到最终分类结果.接下来,将“车流量”、“行车速度”、“道路占用率”、“相邻监测站的车流量差值”、“速度差值”以及“道路占用率差值”等交通参数表示为特征向量,分别输入到经过训练的SVM分类器,并将多SVM分类器融合后的分类结果作为判别异常事件的依据.最后,从5个具有代表性的高速公路路段采集到的交通数据构造实验数据集.实验结果表明,对比单一SVM和LS-SVM,文章提出的基于多SVM分类器融合的高速公路异常事件检测方法可以有效提高高速公路异常事件检测的准确性和可靠性,弥补了仅使用单一交通参数进行异常事件检测的不足.

Expressway Abnormal Events Detecting Method Based on Multi-SVM Classifier Fusion

This paper proposed a multi-SVM classifier expressway abnormal event detection method through complex expressway traffic parameters integration. Firstly, the initial training sets were divided into non-overlapping subsets, in which each subset was used to train a related classifier. Given an input vector, the category labels could be obtained by the related classifier, and then membership degree of the specific vector to a particularly cluster was calculated. Secondly, the multiple SVM classifiers integration results ware achieved by probabilistic methods, and then the final classification results were obtained. Next, six expressway traffic parameters （such as ＂traffic＂, ＂traffic speed＂, ＂road occupancy rate＂, ＂difference between adjacent stations traffic＂, ＂speed difference＂ and ＂road occupancy rate difference＂） were utilized as input eigenvectors to a trained SVM classifier. Furthermore, the abnormal events could be recognized through the muhi-SVM classifier classification results. Finally, performance evaluations were conducted from five representative expressway＇ s vehicle data. Experimental results showed that, compared to single SVM and LS-SVM, the proposed muhi-SVM classifier fusion algorithm based expressway abnormal event detection method could effectively improve the accuracy and reliability for expressway abnormal event detection, and could solve the problem in abnormal event detection only using single traffic parameter.