基于YOLOv5+DeepSORT检测数据的车头时距混合分布模型研究

Research on Time Headway Mixture Distribution Model Based on YOLOv5+DeepSORT Detection Data

中文关键词:  图像检测  车头时距  混合分布模型  目标跟踪

英文关键词:image detection  time headway  mixture distribution model  target tracking

中文摘要:

      为提高车头时距分布模型的准确性，提出一种基于YOLOv5+DeepSORT算法的样本采集方法，并拟合出一种双高斯-移位负指数混合分布模型。训练YOLOv5检测器模型和DeepSORT算法中的ReID模型对车头时距进行自动精确采集，建立双高斯-移位负指数混合分布模型来描述实际交通流中3种车辆驾驶状态：强跟驰、弱跟驰和自由流，对自动采集的车头时距样本进行拟合，利用最大期望算法对模型参数进行标定。经实例验证表明，基于YOLOv5+DeepSORT方法采集的车头时距样本与视频人工逐帧记录的样本之间的平均相对误差为1.94%，满足车头时距样本采集准确率的要求；所提出的双高斯-移位负指数混合模型对自动采集车头时距样本的拟合结果通过了K-S检验，且拟合结果优于三元混合分布、二元混合分布和威布尔分布模型。

英文摘要:

      In order to improve the accuracy of time headway distribution model,a sample collection method based on YOLOv5+DeepSORT algorithm is proposed,and a double Gaussian-shift negative exponential mixture distribution model is fitted.The YOLOv5 detector model and the ReID model of DeepSORT algorithm are trained to collect the time headway automatically and accurately,and a double Gaussian-shift negative exponential mixture distribution model is established to describe three vehicle driving states in real traffic flow: strong following,weak following and free flow.The automatically collected time headway samples are fitted and the model parameters are calibrated by using the maximum expectation algorithm.The results show that the average relative error between the time headway samples collected based on YOLOv5+DeepSORT method and those recorded manually frame by frame is 1.94%,which satisfies the requirement of time headway sample collection accuracy.The fitting results of the proposed double Gaussian shift negative exponential mixture model for automatically collecting headway samples pass the K-S test,and the fitting results are better than the ternary mixture distribution,binary mixture distribution and Weibull distribution models.