一种识别重叠噪声的交通量检测算法

Traffic detection algorithm for identify overlapping noise

在传统的交通量检测中,短时能量法和短时平均幅度法存在无法识别重叠行车噪声段的问题,而提取与重叠信号相关的频谱视图(spectrum view,SV)作为重叠行车噪声段的特征有助于解决以上问题。针对以上问题进行研究,提出了一种基于SV特征的交通量检测算法。该算法首先对含有重叠行车噪声段的交通噪声进行滤波降噪;然后将交通噪声通过快速傅里叶变换(fast Fourier transform,FFT)以提取其SV特征,并对SV特征进行平滑处理;最后基于SV特征并以双门限判决原理为依托,对交通噪声进行端点检测并分离出重叠的行车噪声段。基于单车道、少流量路段的交通噪声数据集的实验表明,与传统方法相比,本文算法准确率提高了20%,从而验证了该算法的有效性。

Traditionally,the short-term energy method or the short-term average magnitude method is not able to identify the overlapping traffic noise segments effectively.In order to solve the problem,this paper proposed a traffic detection algorithm based on the SV features.First,it filtered and denoised the traffic noise which contained overlapping noise.Next,it extracted the SV features of the traffic noise via FFT,and then smoothed the SV features.Finally,based on the SV features and the principle of dual-thresholds judgment,it detected the endpoints of the traffic noise and identified the overlapping noise.In condition of a single lane and low traffic,experiments on dataset of traffic noise demonstrate that the accuracy outperforms other usual methods by 20%.