Pedestrian Injury Prediction In Pedestrian-Vehicle Conflict Based On Probabilistic Method

The uncertainties in pedestrian mobility are important factors for risk assessment of pedestrian collision avoidance system.In this paper,a novel real-time method to predict the pedetrian injured probabilit`ies is provided by counting the uncertainties of pedetrian motion.Time metrics and collision probabilities are the major measurements for risk assessement now,but the first one simplifies the pedestrian mobility as deterministic model and the second one ignores the effect of collision speed.So it is not appropriate that time metrics and collision probabilities become the decisio n making basis for collision avoidance system.Besides,in probabilistic approaches,pedestrian model and distance model are important prerequisites for risk assessment,and a reasonable pedestrian model can improve the robustness of system.While the existing distance models either exaggerate the risk or can not distinguish the frontal impact from side ones.Acctually,side impact for pedestrian-vehicle rarely occurs.Also,prediction method should be protential for real-time application.Based on four aspects: pedestrian model,distance model,risk assessment description and prediction method,this paper was established as follows:(1)Build a stochastic pedestrian model corresponding with the motion of real pedestrians.The first-order Markov process was used to desribe the uncertainties in pedestrian motion,and the real data of pedestrians crossing road in Changsha city was recorded to acquire the parameters in pedestrian model.Then,to verify the validity of stochastic model,we compared the modeling trajectories with real trajectories.(2)Present a confilct distance model only counting the frontal collisions.This distance model can exactly reflect the threat pedestrian facing in the process of conflict,where the conflict distance was obtained through two scebarios by counting the relative motion of pedestrian and vehicle.(3)Propose the pedestrian injured probabilities to describe the conflict risks.Countiong collision speed and collision probability which are important factors for risk prediction,the injured probabilities of pedestrian are proposed as the conflict risk to improve the robustness of system decision,which were obtained on the basis of the stochastic pedestrian model and conflict distance model.(4)Predict the pedestrian injured probabilities based on the design and application of Unscented Transformation.Target probability density functions conforming toGauss distribution were proved by Monte Carlo random sampling.After that,to improve the computational efficiency of system,the symmetric Unscented Transformation was used to predict the injured probabilities.Then the presented distance model and method assessing injured probabilities are verified with Monte Carlo(MC)as the reference in 20000 stochastic pedestrian-vehicle conflict scenarios,and show the high accuracy and potential for real-time application.To evaluate the effectiveness of the probabilistic injury assessment,a pedestrianvehicle conflict scenario is constructed.Simulations were performed to assess the probability of injury to pedestrian with different vehicle maneuvers adopted to avoid collision.Comparing to the deterministic method,the new method shows more accurate prediction performance,and provides an effective quantization means for collisionavoidance decision and control.