Abstract
Road traffic monitoring is very important for intelligent transportation. The detection of traffic state based on acoustic information is a new research direction. A vehicles acoustic event classification algorithm based on sparse autoencoder is proposed to analysis the traffic state. Firstly, the multidimensional Mel-cepstrum features and energy features are extracted to form a feature vector of 125 features; Secondly, based on the computed features, the five-layers autoencoder is trained. Finally, vehicle audio samples are collected and the trained autoencoder is tested. The experimental results show that detection rate of the traffic acoustic event reaches 94.9%, which is 12.3% higher than that of the traditional Convolutional Neural Networks (CNN) algorithm.
Highlights
In order to provide auxiliary research for intelligent transportation and traffic safety development, an autoencoder based the acoustic feature for traffic event detection is proposed
In order to integrate the dynamic features of the sound, the algorithm extracts the multidimensional Mel-cepstrum features and energy features, and forms a 110 dimension feature vector
The experimental results show that the detection rate of traffic acoustic events reaches 94.9%, and the recognition rate of collision sounds reaches 97. 9%
Summary
In order to provide auxiliary research for intelligent transportation and traffic safety development, an autoencoder based the acoustic feature for traffic event detection is proposed. The experimental results show that the detection rate of traffic acoustic events reaches 94.9%, and the recognition rate of collision sounds reaches 97. The detection of traffic state based on acoustic information has been an important research direction for intelligent transportation. The specific recognition model based the sound signal was constructed for four different vehicles, and the recognition accuracy was 73.68%. Some scholars have tried to apply convolutional neural networks (CNN) to recognize sound event[4]. Convolutional neural networks have greatly improved recognition rate and recognition speed. There are still some problems when the traditional CNN model is applied to sound event recognition.
Sign-in/Register to view highlights & summary 
Published Version (
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