Huang Zhi, Liu Xiangyi, Song Xiaolin, He Yin
a College of Mechanical and Vehicle Engineering , Hunan University , Yuelu District, Changsha , P.R.C.
b State Key Laboratory of Advanced Design and Manufacturing for Vehicle Body, Hunan University , Yuelu District, Changsha , P.R.C.
Traffic Inj Prev. 2017 Aug 18;18(6):650-656. doi: 10.1080/15389588.2017.1283026. Epub 2017 Jan 23.
The uncertainties of pedestrian mobility are important factors affecting the accuracy and robustness of an active pedestrian protection system. This study is to provide the means for probabilistic risk evaluation of pedestrian-vehicle collision by counting the uncertainties in pedestrian motion.
The pedestrian is modeled by a first-order Markov model to characterize the stochastic properties in mobility according to field experiments of pedestrians crossing an uncontrolled road. Based on the assumption of Gaussian distribution, unscented transformation (UT) is employed to predict the collision risk probability with the symmetric σ-set constructed on the basis of discrete trajectory simulation. Simulation experiments were carried out with 10,000 Monte Carlo (MC) simulations as the reference.
The probability density distributions of time-to-collision, minimal distance, and collision probability estimated by UT coincide with the reference ones under various vehicle-pedestrian conflict scenarios, and the maximal deviation of collision probability from the reference is 5.33%. The UT method is about 600 times faster than the MC method (10,000 runs), which means that the proposed method has the potential for online application.
This article presents an effective and efficient algorithm to estimate the collision probability by using a UT method to solve the nonlinear transformation of uncertainties in pedestrian motion. Simulation results show that the UT-based method achieves accurate collision probability estimation and higher computation efficiency than MC and provides more valuable information concerning collision avoidance than the deterministic methods in the design of a pedestrian collision avoidance system.
行人移动性的不确定性是影响主动式行人保护系统准确性和鲁棒性的重要因素。本研究旨在通过计算行人运动中的不确定性，为行人与车辆碰撞的概率风险评估提供方法。
根据行人穿越无控制道路的现场实验，采用一阶马尔可夫模型对行人进行建模，以表征其移动性中的随机特性。基于高斯分布假设，采用无迹变换（UT），通过基于离散轨迹模拟构建的对称σ集来预测碰撞风险概率。以10000次蒙特卡罗（MC）模拟作为参考进行仿真实验。
在各种车辆与行人冲突场景下，UT估计的碰撞时间、最小距离和碰撞概率的概率密度分布与参考分布一致，碰撞概率与参考值的最大偏差为5.33%。UT方法比MC方法（10000次运行）快约600倍，这意味着所提出的方法具有在线应用的潜力。
本文提出了一种有效且高效的算法，通过使用UT方法解决行人运动中不确定性的非线性变换来估计碰撞概率。仿真结果表明，基于UT的方法实现了准确的碰撞概率估计，且比MC具有更高的计算效率，在行人碰撞避免系统设计中比确定性方法提供了更有价值的碰撞避免信息。
