Application and Numerical Investigation of Wavelet Transform for Traffic Data Denoising

Good fitting of traffic data is important to traffic study because initial and boundary conditions of dynamic traffic models and relationships among traffic variables are dependent on the calibration of data. In this paper, a denoising method of traffic data, such as speed, density and flow, is proposed and discussed numerically. The denoising procedure is based on Daubechies wavelet transform and soft thresholding method. An empirical study with eight Daubechies wavelet transforms and five degree of smoothing are discussed and compared. We find db8 with 80 % degree of smoothing presents the best denoising result. After denoising, r-square value of speed-density fitting is improved from 0.863 to 0.956. Key-Words: Wavelet transform, data preprocessing, Daubechies wavelets, denoise, speed-density relation.