基于随机森林的车辆行为分析方法

Analysis of vehicle behavior based on random forest

为了实现智能交通领域异常车辆的自动识别,结合实际情况及大量的实验仿真,提出一种由目标车辆检测跟踪、车辆轨迹处理、车辆行为分析三部分组成的解决方案。首先采用改进的高斯模型背景差分算法来检测目标车辆,接着采用基于粒子滤波算法跟踪目标车辆,然后拟合跟踪轨迹以提取特征值,最后将特征样本输入随机森林算法生成分类器,对车辆行为进行分析。实验结果表明该方法能够实现对现实场景中车速过快和频繁变道两种异常行为的识别。

In order to realize abnormal vehicle be identified automatically in ITS（ Intelligent Transportation）. Combined with the actual situation and a lot of experimental simulation,this paper puts forward a solution three parts,including target vehicle detection and tracking,vehicle trajectory processing and vehicle behavior analysis. Firstly,adopt modified algorithm of gaussian model background different to detect the target vehicle. Then the vehicle is tracked by particle filter. Then the tracking trajectory is fitted to extract the eigenvalue. Finally,the feature sample is entered into the random forest to generate classifier,for analysis of vehicle behavior. The experimental results show that this method can realize two abnormal behaviors of high speed and frequent change in the real situation.