Y. Gao, J. Fu, W. Feng, K. Yang

Pages: 299-314

Abstract
The Next Generation Simulation (NGSIM) vehicle trajectory dataset is a significant public dataset in the field of transportation. However, the presence of outliers in the dataset due to measurement errors, environmental influences, and data transmission errors undermines the reliability of research based on this dataset. To address this issue, this study proposes a vehicle trajectory reconstruction method for NGSIM based on the isolation forest algorithm, Gaussian kernel regression algorithm, and wavelet analysis. Firstly, the isolation forest algorithm is employed to detect outliers. Secondly, the Gaussian kernel regression algorithm is utilized to correct these outliers. Lastly, wavelet analysis is applied for denoising purposes. Through experimental analysis of the NGSIM vehicle trajectory dataset I-80, the results demonstrate that using Sym3 as the basis function yields optimal denoising effects whether by correcting original velocity data with Gaussian kernel algorithms followed by wavelet denoising or directly applying wavelet denoising alone. The signal to noise ratio (SNR) and root mean square error (RMSE) values obtained from correcting original data with Gaussian kernel algorithms followed by wavelet denoising are 24.02 and 1.40 respectively. These values indicate higher SNR and lower RMSE compared to directly applying wavelet denoising on original data alone. Wavelet analysis exhibits superior denoising effects when applied after correcting original data with Gaussian kernel algorithms while preserving relevant characteristics of the original data.
Keywords: NGSIM; vehicle trajectory reconstruction; isolation forest algorithm; gaussian kernel regression algorithm; wavelet denoising