基于路侧毫米波雷达的车辆碰撞概率计算方法

为定量化得出高速公路同一车道中前后相邻车辆的碰撞概率，从制动减速度的角度出发， 提出一种新的前后相邻车辆碰撞概率计算方法。分别考虑前后车发生碰撞的3种不同情况，推导出如果发生碰撞前车需要的最小制动减速度。基于路侧毫米波雷达获取海量车辆运行状态真实数据，包括轨迹、速度以及制动减速度的变化规律，采用广义帕累托分布(Generalized Pareto Distribution，GPD)建立制动减速度分布模型，进一步基于GPD模型计算出在不同场景下如果发 生碰撞所需最小制动减速度的发生概率，将该概率值确定为碰撞概率。研究结果表明，在本研究路段，约99.10%的加速度在-1, 1] m·s-2 的区间范围内波动，车辆制动减速度的分布具有“长尾” 特征，较大的制动减速度占比非常小。内侧1车道、2车道加速分布比3车道的分布更为集中，大型货车的加速度分布比小客车的加速度分布更集中。最后，基于真实的危险场景数据以及模拟的典型危险场景数据进行验证，将该方法的计算结果与传统方法的计算结果相比较，表明该方法的计算结果连续，且可迅速、准确地识别各类危险场景。

A Calculation Method of Vehicle Collision Probability Based on Roadside Millimeter Wave Radar

The purpose of this research is to study the collision risk of the following vehicles in the same lane on the freeway. A new method for calculating the collision probability between leading and following vehicles was proposed by vehicle deceleration. The minimum deceleration speed required by the leading vehicle was deduced by considering three different collision cases. Furthermore, a large number of real vehicles driving state data were obtained based on roadside millimeter- wave radar. Trajectory, velocity, and deceleration were included. The Generalized Pareto Distribution (GPD) was used to develop the distribution model of deceleration. Based on the GPD model, the occurrence probability of minimum deceleration was calculated, and the probability was determined as the collision probability. The results showed that about 99.10% of vehicle acceleration fluctuated within the range of -1, 1] m·s-2 . The acceleration of most vehicles in this study section was very small. The distribution of deceleration showed the feature of a "long tail". The proportion of large braking deceleration was very small. The acceleration distribution of lane 1 and lane 2 on the inside was more concentrated than that of lane 3. The acceleration distribution of large trucks was more concentrated than that of small vehicles. Furthermore, based on the actual and simulated dangerous scene data, the calculation results of the proposed method were compared with those of the traditional method. It is shown that the calculation results of this method are continuous, especially it can quickly identify all kinds of dangerous scenes.