Design of algorithm for vehicle identification by number plate recognition

Abstract

Design of an algorithm for vehicle identification by recognizing the number plate is presented. This new vehicle identification technique consists of vehicle detection, plate localization and character recognition. Here, Genetic algorithm (GA) is employed at two levels: for detecting vehicle from traffic image and recognizing character from the number plate. Detection is based on contour and shape information. GA controls the window size to capture each vehicle in a separate widow. Connectivity and adjacency concepts are used to locate and extract number plate and its characters. A digital board (DB) with window panes is introduced to recognize each character uniquely. GA is adopted at the second level to map character pixels into the window panes as lines. For each character in the number plate, distinct feature vector is computed. Finally, a feature based matching is adopted for character recognition. Experiments have been conducted with images taken from various scenes and conditions and the detection rate is found to be 92.5 %. Experiments have conducted for recognition with LPR images taken at different conditions and the recognition rate is found to be 91 %. Detection time is linear function of number of objects in the input image. Potential applications include provisioning of vehicle parking facilities and campus security system for permitting authorized vehicles into the premises.