Vehicle Road Rage Driving State Recognition Method Based On Speech And Text Fusion

Road rage is an important factor affecting driving safety.Road rage driving identification and real-time warning are of great significance for reducing traffic accidents and ensuring road traffic safety.There are few researches focusing on road rage based on speech and text.What’s more,speech and text are two important ways to investigate road rage driving.Based on the analysis of speech and text features in road rage,a road rage driving identification method based on speech and text fusion decision is proposed in this thesis.The main research contents of this thesis are as follows:Firstly,constructed data acquisition software,obtained speech and text data by designing and organizing simulation and real vehicle driving experiments.The subjective auditory evaluation method was used to annotate the data,and a corpus database suitable for the study of road rage was obtained.Secondly,in the aspect of speech,after the preprocessing of speech data,the key features of the hyper linguistic features,spectral correlation features and voice quality features were analyzed,extracted and fused to construct the speech feature data set.In terms of text,the word frequency in text data was counted by the term frequencyinverse document frequency（TF-IDF）algorithm after the text data was preprocessed.Constructed a sentiment dictionary,optimized text data by using sentiment dictionary,and then constructed a text feature data set.Finally,based on the speech feature data set,firefly algorithm（FA）was used to optimize the probabilistic neural network（PNN）,and the FA-PNN road rage driving identification model based on speech was obtained.Based on the text feature data set,a text-based long short-term memory（LSTM）road rage driving identification model was obtained by using LSTM.The above two models were used to obtain the category probability,and the road rage driving identification model based on speech and text fusion decision was constructed by using the adaptive weight method at the multimodal decision level,and then the experimental data was used to empirically analyze the three models.The results indicate that the road rage driving identification model based on speech and text fusion decision has the highest accuracy of 94%,4% higher than FA-PNN and8.5% higher than LSTM.This thesis demonstrates the feasibility of speech and text identification for road rage driving,which provides a new method for the study of road rage driving and advanced driver assistance systems.