Abstract
Traffic state information is widely applied into all aspects of Intelligent Transportation System (ITS), such as the macro-control of government departments, the implementation of traffic managers' plans, the decision-making of residents travel, and so on. At present, Mel Frequency Cepstrum Coefficient (MFCC) is generally used as characteristic of traffic noise to characterize different traffic states, and performs well in simple noise environment, but performs poorly in complex noise environment. Based on the analysis of traffic noise acquired from a roadside-installed acoustic acquisition equipment, the evaluation problem of traffic state in complex noise environment is considered in this paper. Traffic state is divided into three categories according to traffic speed in our work: free flow (40 km/h and above), saturated flow (10-40 km/h), and jammed flow (0-10 km/h). Teager Energy Operator (TEO) is introduced to improve the MFCC characteristic, thus a novel characteristic called T-MFCC is proposed. Principal Component Analysis (PCA) is introduced to reduce dimension of T-MFCC characteristic, thus a novel characteristic called PT-MFCC is proposed. Support Vector Machine (SVM) optimized by Particle Swarm Optimization (PSO) algorithm is applied as classifier to identify traffic state. Characterization capabilities of two modified characteristics and traditional MFCC characteristic for traffic state are compared in this paper. Experimental results demonstrate that the evaluation accuracy of traffic state based on T-MFCC characteristic is 3.685% higher than that based on MFCC characteristic, and the evaluation accuracy of traffic state based on PT-MFCC characteristic is 26.466% lower than that based on MFCC characteristic. Therefore, T-MFCC characteristic is superior to MFCC characteristic, while MFCC characteristic is superior to PT-MFCC characteristic, namely, T-MFCC characteristic can better characterize traffic state than MFCC characteristic, meanwhile, there are no redundancy attributes in T-MFCC characteristic, thus PCA is not needed to reduce the dimension of T-MFCC characteristic.
Highlights
Statistics from China Statistical Yearbook and the Ministry of Public Security show that from 2005 to 2019, the ownership of civil vehicles increases from 16.1835 million to 220 million, and the ownership of private vehicles increases from 13.544 million to 207 million
In literature [18], a traffic state evaluation method based on Mel Frequency Cepstrum Coefficient (MFCC) characteristic of traffic noise is proposed: First, the traffic state is still divided into free flow (40 km/h and above), saturated flow (10-40 km/h), and jammed flow (0-10 km/h); Second, the traffic noise data is still collected from three different places; Third, the MFCC characteristic of traffic noise is extracted, the Gaussian Mixture Model (GMM) and Support Vector Machine (SVM) are utilized as classifier to identify the traffic state, and the SVM classifier is implemented by LIBSVM toolbox; the experimental results show that SVM owns higher evaluation accuracy than GMM
Compared with PT-MFCC characteristic, the recognition rate of free flow corresponding to T-MFCC characteristic is increased by 41.708%, the recognition rate of saturated flow is increased by 49.748%, the recognition rate of jammed flow is reduced by 1.005%, and the comprehensive evaluation accuracy is improved by 30.151%
Summary
Statistics from China Statistical Yearbook and the Ministry of Public Security show that from 2005 to 2019, the ownership of civil vehicles (small passenger vehicles) increases from 16.1835 million to 220 million, and the ownership of private vehicles (small passenger vehicles and miniature passenger vehicles) increases from 13.544 million to 207 million. In literature [18], a traffic state evaluation method based on MFCC characteristic of traffic noise is proposed: First, the traffic state is still divided into free flow (40 km/h and above), saturated flow (10-40 km/h), and jammed flow (0-10 km/h); Second, the traffic noise data is still collected from three different places (in city); Third, the MFCC characteristic of traffic noise is extracted, the Gaussian Mixture Model (GMM) and SVM are utilized as classifier to identify the traffic state, and the SVM classifier is implemented by LIBSVM toolbox; the experimental results show that SVM owns higher evaluation accuracy than GMM. Literature [10] conducts an 11.5-day experimental study on the roadside of Paris Ring Road, extracts the MFCC characteristic of traffic noise and estimates the traffic volume with the Support Vector Regression (SVR) method, the experimental results demonstrate a good application prospect.
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Free) 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
