Acoustic Signal Processing Based Automatic Detection Method For Traffic Accident

Traffic injuries have become a serious issue to the safety of human lives, it's of great practical significance to establish an emergency rescue system that is fast and efficient, so as to change the traditional Rescue mode. However, the traditional traffic incident detection approaches focuses on macroscopic traffic flow information, paying little attention on microscopic phenomena like traffic accident, resulting in unsatisfactory accuracy of real-time traffic accident detection.In this paper, acoustic signal processing based automatic detection method for traffic accident is studied, the detection algorithm runs on the DSP board, by collecting and analyzing the sound around the vehicle, and real-time monitoring is carried out. When a traffic accident is detected, the ambulance system far away will receive an alarm signals sent by the DSP system.This paper uses the pattern recognition method to analysis the acoustic signals collected, including three main stages: preprocessing, feature extraction and real-time classification. Classifier parameters are obtained by off-line training process. The haar wavelet transform is used to extract frequency domain information, while the outliner detection method one-class support vector machine is used for classification. Thus, it overcomes the limitations of linear discrimination and maintains a good generalization ability. Compared with common LDA algorithm, it has higher classification accuracy. The proposed algorithm also has the advantages of lower complicated computing and easy to implement. DSP-based traffic accident detection system is designed. In the DSP system, varies of identification information is integrated into the classification algorithm, which ensures both real-time efficiency and accurate property of the designed system. Both simulation platform and practical on-site experimental results indicate that this algorithm achieves satisfactory performance, outperforming most of current approaches. The results of simulation experiment show that this algorithm can be used in a practical accident detection system.