Automatic License Plate Recognition Using K-Nearest Neighbour Approach

Licence plate detection and recognition (LPDR) is difficult problem which is essential intelligent transportation systems (ITS). This could be employed as a core component in variety of elements, including traffic control, electronic payment, and security systems. For this work, different algorithms are created, each with benefits and limitations for extracting images of plate below varying conditions. Nevertheless, due to some approach's complexity, substantial calculation costs may be required, which might time-consuming. This work, proposes K-Nearest Neighbour (KNN) model, which is a simple and effective for detecting licence plates and recognising character characters. The licence plate is recognised initially using 2-D wavelet transformations to extract vertical edges from input image. Vertical edge's high density is estimated initially to identify suitable licence plate areas. The possible locations are then confirmed utilizing a plate/ non-plate KNN classifier. Next LP is recognised, characters are split utilizing straightforward algorithm depend on empty distance among characters. At last, the character candidates are categorised by training another KNN classifier. Implemented KNN method obtains better performance when compared to existing methods such as robust LP recognition method, deep convolutional neuron network, and improved lightweight YOLOv5s model, and result with a high accuracy of 99.69%, 99.89%, and 99.53% using AOLP, and CCPD datasets.