Research On Pedestrian Crossing Intention Recognition Model Facing The Demand Of Vehicle-mounted Pedestrian Warning System

The vehicle-mounted pedestrian early warning system uses sensors to identify pedestrians,alerts the driver when there is a risk of pedestrian-vehicle conflict,or automatically makes emergency braking to avoid accidents.At present,the early warning system for pedestrians has begun to be applied in a certain range,but the early warning algorithm of these systems is simpler overall,and there is a problem of insufficient accuracy and high false positive rate.In order to improve the use of pedestrian early warning system,we must first analyze and predict the pedestrian’s cross-street intention in advance,and realize the accurate perception prediction of pedestrian-vehicle interaction.In view of the above objectives,a signalless zebra line segment was selected as the research object,and the interactive data of pedestrian-vehicle in zebra line segment were collected for a long time using lidar sensors and small cameras,and the motion status data of pedestrian crossing and approaching zebra crossing vehicles in different risk states were obtained.Based on this,the paper analyzes and determines the set of characterization parameters related to pedestrian crossing intention,establishes the model of pedestrian crossing intention by machine learning method,and studies the cross-street behavior characteristics of pedestrians in two states: zebra crossing line.The main research in this paper is as follows:(1)This paper makes a specific analysis of the pedestrian crossing characteristics in pedestrian crossings where there is no signal light to control the pedestrian crossing,and a specific analysis of the characteristic parameters that affect pedestrian crossing decisionmaking(including: pedestrian number,gender,age,speed change and speed and relative distance of cross-street game vehicles).The results show that the sex,age and number of crossstreet groups all cause the difference of pedestrian crossing speed to some extent,the speed of pedestrian crossing is faster for men than for women,the efficiency of young pedestrians crossing the street is higher than that of middle-aged and older people,and the number of pedestrian groups and the waiting time of crossing the street are obviously antilinear.(2)Based on the results of numerical analysis of pedestrian crossing characteristics,a model support vector machine(SVM)classification is established,the sample characteristic parameter data is studied and predicted,and the prediction accuracy of the sample characteristic data is calculated by the SVM classifier.Because the number of characteristic parameters will cause the fluctuation of the accuracy of SVM classification model,this experiment predicts the model of five kinds of feature parameters and six kinds of feature parameters respectively,and concludes that: The SVM classification model has a good prediction effect on pedestrian crossstreet intention classification,and when selecting six types of feature parameters,the average accuracy of SVM classification model’s prediction of sample data is better than that of selecting five types of feature parameters.(3)The pedestrian crossing intention in two cases,with and without pedestrian crossing,is predicted,and it is concluded that the average prediction classification effect of the model of the pedestrian crossing is significantly better than that of the model at the unmarked crosswalk,and the accuracy difference is 1.59%.SVM classification model can better predict the crossstreet intention of pedestrians who choose pedestrian crossings,pedestrians crossing motorways without pedestrian crossings,because their crossing paths are more dispersed,the purpose and direction are more unstable than pedestrians on pedestrian crossings,resulting in more difficult to predict their cross-street intentions,reflected in the accuracy of model prediction for the classification model pedestrian crossing intention prediction accuracy is lower.