| Highway tunnels play an important role in improving highway technical status,shortening running distance and improving transportation capacity.With the increase of the number and total mileage of tunnels,the probability of traffic accidents in tunnels is also increasing.In view of the defects of low efficiency and safety of manual inspection,the tunnel traffic accident perception method is particularly important.Therefore,this paper proposes a tunnel traffic accident perception method based on acoustic signals to explore the feasibility of acoustic signals used in tunnel traffic accident detection.In the sensor-based accident detection research,auditory perception can supplement visual perception,through the fusion of the two information can improve the existing tunnel traffic accident detection system,so as to improve the overall intelligent level and safety of our tunnel construction.First,the main research content of this paper is established.The acoustic perception method proposed is mainly reflected in three aspects,namely,abnormal sound segment detection in the tunnel,abnormal sound recognition and accident severity judgment.In view of the above three contents,this paper analyzes the relevant researches and shortcomings of domestic and foreign researchers in abnormal sound detection,sound recognition and sound quality evaluation.By analyzing the process of feature extraction of tunnel noise signal,the difficulty of collecting actual samples is synthesized and the method of establishing accident noise set in tunnel environment is obtained.Secondly,the existing research methods based on acoustic signals are analyzed and corresponding improvements are proposed.Based on the existing research,the threshold and threshold endpoint detection combined with signal characteristics are used to detect abnormal sound segments.The previous experiments have found the defects of the traditional energy entropy ratio method in signal feature extraction,so the improved energy entropy ratio method is proposed to detect abnormal sound segments in tunnels.At the same time,a new feature,LT-MFCC,is proposed to solve the problem of low recognition rate of Mel frequency cepstral coefficients in sound recognition.By combining this feature with particle swarm optimization support vector machine,abnormal sounds are classified and used for abnormal sound recognition in tunnel environment.Then,a judgment method for tunnel traffic accident severity was established based on acoustic quality evaluation technology.The loudness of psychoacoustic parameters was selected and combined with time-frequency perception weighting to construct the proposed evaluation model SMA.The judgment process of tunnel traffic accident severity based on this model was divided into seven steps.It includes collision noise preprocessing,Mayer basic feature extraction,Mayer feature matrix conversion,collision duration perception definition,frequency domain weighting factor determination,time varying perception definition and SMA evaluation model establishment.Finally,the experimental study of tunnel traffic accident perception based on acoustic signal is carried out.In the abnormal sound segment detection experiment,the traditional energy entropy ratio method and the improved energy entropy ratio method are compared,and the detection accuracy of the proposed method is improved by 3.09%compared with the traditional method.In the abnormal sound recognition experiment,for the five types of abnormal sounds in the tunnel selected,the recognition accuracy of the features constructed in this paper is improved by 19.0%and 28.2%respectively when the signal-to-noise ratio is 5d B and 1d B,compared with the traditional characteristics of Merle’s inverted spectrum coefficient.In the experiment of accident severity judgment,SMA was compared with two traditional time-varying loudness models,and subjective scores were collected to compare the three models.Finally,the proposed evaluation model was more suitable for accident severity evaluation than the two traditional models. |
PDF Full Text Request