基于视频识别的混合非机动车速度分布模型

Speed distribution model of mixed non-motorized vehicles based on video recognition

中国非机动车出行近年来逐渐复兴并形成了复杂的混合交通流动,造成了交通运行和安全问题,因此对混合非机动车速度分布进行准确建模是重要的现实需求。该文首先介绍了利用深度神经网络进行多目标跟踪的数据采集方法,然后输入非机动车道的拍摄视频进行自动识别,高效获取实测数据;在对车速进行统计分析后,通过信息准则确定混合Gauss模型的最优组分,采用期望最大化算法求解模型参数的极大似然估计,并建立模型参数与道路运行状况和统计特征之间的联系。所使用的智能化视频识别方法提高了数据采集效率,求解参数前进行组分选择可以提高建模效率。拟合结果表明:混合Gauss模型对非机动车速度分布的描述比单分布模型更准确。在流动不受阻碍时,混合Gauss模型的结果以非机动车类型划分组分,其参数与速度均值、标准差及各类型车的比例相关;在高峰时,组分划分与根据流动状态进行分类是一致的,其中较快组分的平均速度接近车辆自由流动速度。

Non-motorized vehicle travel has gradually revived in China to form complex,mixed traffic flows in recent years that lead to traffic movement and safety problems.The characteristics of the mixed non-motorized vehicle speeds need to be accurately modeled to characterize this problem.Deep neural networks are used here for multi-target tracking in videos of the non-motorized vehicle lanes to measure the non-motorized vehicle speeds.A statistical analysis of the vehicle speeds is used to determine the number of components of the Gaussian mixture model for the speeds using information criteria with the maximum likelihood estimate of each parameter calculated using the expectation maximization algorithm.Then,the model parameters are related to the road operating conditions and statistical characteristics.This intelligent video recognition method accelerates data collection to obtain sufficient data and the component determination prior to other parameters improves the modeling efficiency.The fitting results show that the Gaussian mixture model more realistically describes the speed distribution of non-motorized vehicles than the single distribution model.The Gaussian mixture model results are divided into the various types of non-motorized vehicles when the vehicle flow is not obstructed and the parameters are related to the average speed,the standard deviation and the ratio of different vehicle types.The classifications according to flow states are consistent during peak periods with the mean speed of the faster component close to the free flow velocity of the vehicles.