Wavelets in intelligent transportation systems: Data compression and incident detection

Research show that wavelets can be used efficiently in denoising and feature extraction of a given signal. This thesis discusses about intelligent transportation systems(ITS), its requirement and benefits. We explore use of wavelets in intelligent transportation systems for knowledge discovery, compression and incident detection. In the first section of thesis, we focus on the following problems related to traffic matrix: data compression, retrieval and visualization. We propose a methodology using wavelet transform for data visualization and compression of traffic data. Aim is to research on the wavelet compression technique for the traffic data, come up with the performance of various available wavelets and the best decomposition level in terms of compression ratio and data distortion. We further investigate use of Embedded Zero Tree (EZW) encoding and Set Partitioning in Hierarchical Trees (SPIHT) algorithm for compression of the traffic data.;In the second section of thesis, we focus on regression model for dichotomous data, i.e. logistic regression. This model is suitable when the outcome can takes only limited number of values, in our case only two, presence or absence of an incident. We look into generalized linear model (gle) with binomial response and logit link function. We present a framework to use logistic regression for incident prediction in transportation systems. Further in the section, we investigate feature extraction using DWT, and effect of preprocessing of data on the performance of incident detection models. A hybrid logistic regression-wavelet model is proposed for traffic incident detection.