基于视频的交通流检测在智能交通系统中具有重要意义。本文针对广泛采用的低位摄像机，提出了一种交通流特性参数的检测分析方法。首先基于三级虚拟检测线和自适应更新率局部背景建模来快速提取车辆特征点并消除活动阴影对提取精度的影响；然后基于Adaboost（Adaptive Boosting,自适应增强）分类器实现特征点按车分组，并在跟踪过程中根据运动特征相关度消除分组误差，获取高精度的车辆轨迹；进而自动生成多车道轨迹时空图并提取各车道交通流的多种特性参数。实验结果验证了算法的高效性；同时，自动生成的多车道轨迹时空图也为更多的交通信息获取和更深入的交通流特性分析提供了有力支持。%Video Based detection of traffic flow has great significance in intelligent transportation systems. For the low angle cameras, a novel traffic flow multi-parameters detection method is proposed in this paper. Three virtual detecting lines and a local background modeling with adaptive learning rate are used to quickly extract vehicle feature points and eliminate the influence of activity shadow. Based on a trained Adaboost (Adaptive Boosting) classifier, the feature points are grouped to vehicles. Then the grouping errors are eliminated based on the motion-similarity of feature points in tracking process and the vehicle trajectories are extracted accurately. After that, the multi-lanes time-space diagrams are generated and the multi-parameters of traffic flow are detected automatically. Experimental results prove the efficiency of the method. In addition, the multi-lanes time-space diagrams can provide strong support for more traffic information acquisition and more in-depth analysis of traffic flow characteristics.
展开▼
