Study On On-board Pedestrian Warning Strategy Based On Hidden Markov Model

At present,some on-board pedestrian early warning systems have been applied in the field of vehicle safety,but the early warning algorithms of these systems are relatively simple as a whole,especially the early warning algorithms are relatively loose,which leads to the high frequency of early warning signals and makes drivers numb and resistant to the early warning signals,and affects the further promotion and use of the system.In this paper,a zebra crossing without signal control is selected,and a large number of natural observation experiments are carried out on pedestrian vehicle interaction conflict data by using lidar and dash cam.The data samples obtained cover more than 20000 vehicles and 5000 pedestrians.Based on the above data,the main research contents and conclusions are as follows:(1)In order to study the behavior characteristics of drivers and pedestrians crossing the street,this paper analyzes the data and concludes that the time between vehicles and zebra crossing is 3s,which is the time threshold for pedestrians to judge whether the current crossing situation is dangerous or not: when the value is less than 3s,pedestrians choose to hover in place;When the value is greater than 4s,pedestrians choose to cross the street directly;When the value is 3-4s,the pedestrian crossing patterns are different: young people choose to cross the street directly,middle-aged and old people choose to linger in the same place,which provides a theoretical basis for the establishment of conflict discrimination rules in this paper;(2)In this paper,the trajectory data is divided into training set and test set according to7:3,and an effective target trajectory prediction model is established by Baum Welch algorithm.Through Viterbi algorithm,state transition probability matrix and divergence probability matrix,the track value of the first 1.44 s of vehicles and pedestrians in the test set is used to predict the track value of the next 0.48 s,and the results are compared with the real results.The results show that: the model is more accurate for the target motion distance prediction,but the prediction of human motion direction shows differences in different types of Cross Street: the prediction of human motion direction is more accurate in the type of direct cross street,and the error is basically below 30°.(3)Because the early warning system has strict requirements for pedestrian position prediction,this paper uses the predicted pedestrian distance and direction to generate the pedestrian coordinate position in the next 0.48 s,and compares it with the real position.The results show that the error between the predicted value generated by the model and the real value is small,which verifies the effectiveness of the trajectory prediction model proposed in this paper.(4)Based on the established target trajectory prediction model,this paper proposes a conflict early warning rule,which defines the conflict area between vehicles and pedestrians,and takes the time when vehicles arrive at the zebra crossing as the threshold to judge whether there is conflict between them.Compared with the real results,it shows that the value of 3S is reasonable,and the recognition rate of conflict is 79.6%,It can accurately predict the potential conflict between the two,and provide a new idea for the on-board pedestrian early warning system.