| Eight new models have been developed and presented in this thesis to process and analyze digitized monochrome and color images. The functions of these models include moving object detection, vehicle signal light detection, noise removal, overall vehicle dimension estimation, and vehicle classification. These models are then integrated to form four algorithms, each dedicated to a specific function: measurement of traffic volume and vehicle speed; detection and count of vehicles intending to turn; classification of vehicles; and measurement of pedestrian flow.;In order to verify the accuracy of the proposed algorithms and their associated software, field studies were carried out using software that has been developed to extract associated traffic data from video tapes. These field studies consisted of approximately ten hours of video tapes which had recorded the actual traffic scenes of several locations in downtown Montreal and a main highway in the Montreal area. Subsequently, these video records were utilized for collection of traffic data pertaining to traffic volume, vehicle speed, vehicle classification and pedestrian volume. Correlation was observed between the traffic data generated by the algorithms and those collected by human observers from a video monitor. The accuracy level obtained in all cases was higher than 90%.;The newly developed models and algorithms provided a new method for increasing the capability and reducing the detection error of currently used video traffic detection systems. With further improvement, the models and algorithms can be used to offer a wide variety of possible applications in intelligent vehicle-highway systems (IVHS). Potential applications include automatic incident detection, automatic queue detection, electronic toll collection, statistical data collection from installed cameras or videotape sequences, and automatic control unit or system for variable-message sign applications in tunnels and on motorways. |
PDF Full Text Request