Traffic Accident Prediction Research Based On Video Similarity SPECIALIZATION:Signal And Information Processing

In recent years,traffic jam and accidents become more frequent,it brings potential problems to our safety and property,seriously affects social stability.Research background conditions of road traffic accident and the cause of accident-prone location,then make the necessary prediction and road safety warning service.It will have huge theory value and practical significance.The thesis is based on the major demonstration project of Jiangsu Provincial Science and Technology Department—the demonstration system based on the freeway traffic sensor network information detection,mining,convergence,publishing and assisting decision.With the analysis of road safety situation at home and abroad,and under the comparative analysis of existing accident prediction-methods and models,adding the further study of the traffic monitoring video information processing technology,we proposed a new road traffic accident prediction method:with video similarity calculation,we can obtain the public attribute information of the similar video set,then analyze the general characteristics of accidents,including the accident time,weather conditions,vehicle types,accident-prone location and so on.It can achieve macro-sense road traffic accident prediction and provide an experience means for road safety management.The similarity calculation based on video node:analyze the traditional attributes similar and relationships similar methods,optimize the SimRank algorithm,then get the similarity computation ObjectSim—define the extracted video attribute node as the general node type,using a weighted rule to distinguish the effects,and calculate the whole relationship graph with the symmetric random walk algorithm.By constructing and analyzing the video similarity prediction model and its corresponding network model,it can lay a solid theoretical foundation for accident prediction research.Road traffic accident prediction based on video similarity:use machine learning,artificial intelligence,data mining techniques to further analyze and process the road accident captured video,extract the value information of video,acquire the similar set with high reliability and accuracy to be used for the subsequent accident forecast analysis.In conclusion,The thesis research results have important theoretical significance and application value.The major contributions of this thesis are as follows:1)Propose a new accident prediction method,with monitoring video similarity calculation,mining the public information of the similar video set,analyze to get the background conditions of accident,achieve macro-sense road traffic accident prediction.2)Optimize the SimRank algorithm,then get the similarity computation ObjectSim,combine the attribute and relationship information of the nodes,and add the effect of different weight value,enhance the accuracy of the similarity calculation.3)Build concrete accident prediction model and analyze its computation process,the related experiments suggest ObjectSim method based on video similarity calculation has the prospective,rationality and feasibility on traffic accident prediction application.