摘要
传统的交通状态检测方法对全景视频中的车辆检测时存在检测精度低、鲁棒性差等缺点。为了解决这些问题,该文提出了一种新的基于虚拟检测线的车辆检测方法。首先,利用提出的基于动态学习率的改进混合Gauss模型构建背景,背景模型的学习率由检测到的车速决定;其次,通过引入Mahalanobis距离来判断虚拟线上的像素是否属于背景;最后,通过设置检测跟踪区域检测车速并跟踪车辆行驶轨迹,避免重复计算车辆数。实验结果验证了所提方法的有效性及在各种场景下较强的鲁棒性。
Traditional traffic state detection approaches for detecting vehicles in full video scenes have drawbacks,such as low detection accuracy and lack of robustness.This paper presents a more effective traffic state detection method based on virtual detection lines using an improved Gaussian mixture model with a dynamic learning rate based on the detected vehicle speed to construct the background scene.The Mahalanobis distance is then used to judge whether the pixels in the virtual detection line belong to the background.Finally,a trajectory zone is selected to obtain the vehicle speed and track its trajectory,while avoiding repeated counting of vehicles.Tests show the method is effective robustness in various situations.
出处
《清华大学学报（自然科学版）》
EI
CAS
CSCD
北大核心
2011年第1期30-35,共6页
Journal of Tsinghua University(Science and Technology)
基金
国家"十一五"科技支撑计划项目(2007BAK12B15)
