This dissertation, "Image Matching of Running Vehicles" by Yin, Lao, 劉然, was obtained from The University of Hong Kong (Pokfulam, Hong Kong) and is being sold pursuant to Creative Commons: Attribution 3.0 Hong Kong License. The content of this dissertation has not been altered in any way. We have altered the formatting in order to facilitate the ease of printing and reading of the dissertation. All rights not granted by the above license are retained by the author. Abstract: Abstract of thesis entitled "Image Matching of Running Vehicles" Submitted by Lao Yin for the degree of Master of Philosophy at The University of Hong Kong in September 2004 This thesis describes a machine vision system that is able to match running vehicles based on their color and license plate information, in order to re-identify (or match) running vehicles passing through two different camera sites by comparing their visual features. The results of matching provide useful information for advanced traffic functions, such as the calculation of link travel time between two different points of a road network and the enforcement of speed limits along a road interval. We suggest solutions to several common problems in visual traffic monitoring systems including the problem of extracting running vehicles under time-varying illuminations and matching vehicles under different illumination conditions; matching running vehicles which may have blurry license marks; and extracting and matching license plates under perspective distortions? The matching process is divided into 4 modules, and novel approaches are suggested for each module. First nodule, named vehicle detection, aims to segment the vehicles from their image sequences. A RANSAC approach is suggested to estimate the background image and thereby to segment the moving vehicles by the background suppression technique. Second module, named shadow removal, aims to remove vehicles' shadows so as to reduce the shape distortion and color distortion. We propose a novel shadow removal method which uses edge information to classify a vehicle image into the shadow region or the object region. Third module, named vehicle classification, aims to classify vehicles into color groups with the aim of increasing the accuracy of the later matching process. We propose the use of k- means clustering with color histogram under some illumination invariant color models. Fourth module, named license plate detection and matching, aims to locate license plates in the image of vehicles and match them based on their visual features. We suggest a method which extracts license plates both color and character texture. Two methods which are based on model fitting and orientation reconstructions are proposed to reduce the perspective distortion. Feature extraction from the normalized license plates is carried out by calculating the topology of the characters. The final matches are found by the correlation of the features vectors. Experiments were conducted outdoors in different weather conditions with two cameras with different viewing angles. The matching rate was 74.2% with zero false alarms, a higher matching rate than that achieved by other matching algorithms in the literature. The results of the experiments also demonstrate that the proposed algorithm is robust to illumination variations and viewing angle variations. To our knowledge, this is the first study to perform vehicle matching across camera sites with vehicle classifications. Conventional license plate methods work with stationary vehicles at car parks or toll gates. This system works with running vehicles on highways. The system is regarded as a superior speed estimator since it estimates the link speed, unlike radars that estimate point speed. (466 words) DOI: 10.5353/th_b3027880 Subjects: Computer vision Image processing Remote sensing Motor vehicles