Construction And Application Of Traffic Conflict Model At Unsignalized Intersection Based On Extreme Value Theory

Non-signalized intersections are an important part of road systems,which are currently widely used on suburban,rural and roads with less traffic.Different from conventional intersections,unsignaled intersections generally have complex and disordered traffic flow and many conflict points,which are more prone to conflict accidents.Therefore,it is very important to construct a traffic conflict model of unsignalized intersection and explore its application fields.In order to explore the conflict characteristics of non-signalized intersections and improve the traffic safety of non-signaled intersections,this paper uses image recognition as a technical means to select and collect conflict data of two non-signalized intersections with different characteristics.Construct single-index block,single-index over-threshold and dualindex over-threshold models for each intersection,and evaluate the pros and cons of model construction,select appropriate conflict index models,and introduce neural networks to evaluate intersection safety.The conclusions of the study are as follows:(1)The UAV video extraction technology is introduced into the intersection conflict recognition,and the intersection traffic flow data is extracted by using the image recognition technology.The results show that,compared with the traditional video extraction technology,this method can effectively eliminate the error caused by video distortion,avoid the correction work caused by the phenomenon of far small and near large,reduce the burden of image recognition,and improve the accuracy of recognition.(2)Select widely used conflict indicators to construct single-index block and overthreshold extreme value models,compare and select parameter models and calculate the estimation results.The results show that the single-index over-threshold extreme value model is significantly better than the single-index block extreme value model in the estimation of small samples;the over-threshold extreme value model also shows better fitting in data fitting.(3)On the basis of the single-index over-threshold extremum model,further explore the construction of the dual-indicator over-threshold extremum model,and also conduct model comparison and result estimation.The results show that the dual-indicator over-threshold extreme value model considers two conflict indicators at the same time,and the fitting result is better,because it avoids pseudo-conflict events where one conflict indicator is large and the other conflict indicator is small,and it is more responsive.Real conflict situations at intersections.(4)By comparing the data utilization,reliability and accuracy of the extreme value model,it is found that the extreme value model beyond the threshold is better than the extreme value model of the block;in the collected conflict indicators of the single-index model,the distance to collision time The result is the best;in the conflict index of the dual-index model,the result of time to collision-post encroachment time shows the best results.(5)For the single-index extreme value model,the neural network algorithm is introduced to calculate and evaluate the safety of extreme value indicators.The results show that in the case of complex indicators,the neural network can autonomously calculate and give the safety evaluation of the intersection.For the dual-index model,the conflict rate index of 10,000 vehicles is used for safety evaluation.The results show that both models can be used for intersection safety evaluation under the condition of selecting the conflict index with better fit,and have strong practical application significance.