Automatic Vehicle Identification Through Visual Features

Detection and recognition of a vehicle license plate is a fundamental requirement of any intelligent transport system, primarily to support activities like finding a stolen vehicle, vehicle surveillance/tracking, parking-toll collection, traffic flow planning and management, etc. However, a license plate can easily be stolen and/or changed by those with criminal intent to conceal their identity. This paper proposes a new vehicle identification system to obtain high degree of accuracy and success rate by not only considering the license plate but also shape of the vehicle. The proposed system is based on four steps: license plate detection, license plate recognition, license plate jurisdiction (province) detection and the vehicle shape detection. In the proposed system, the features are converted into local binary pattern (LBP) and Histogram of Oriented Gradients (HOG) as training dataset. To obtain high degree of accuracy in real-time application, a novel method based on cascaded classifiers is used to update the system. The proposed system allows us to store features of vehicles and related information in the database, thus, allowing us to automatically detect any discrepancy between a license plate and vehicle associated with it.