Number plate localisation and recognition technology based on colour and texture features

To address the problem of significant performance degradation in number plate localisation and recognition from UAV aerial videos due to interference from harsh weather conditions such as rain and fog, this study proposes a technical framework integrating improved image enhancement and robust colour-texture feature analysis. Firstly, a hybrid enhancement model based on Multi-Scale Retinex and Guided Filtering optimized CLAHE is constructed, effectively improving the contrast and detail of low-illumination, rain- and fog-occluded images. In the localisation stage, a robust candidate region extraction mechanism is built by fusing an HSV colour space probability model with improved Uniform Local Binary Pattern texture features, followed by precise localisation using morphological operations and geometric constraints. For the recognition stage, an improved projection feature segmentation method combined with a dual recognition strategy employing SVM and template matching is adopted. Simulation experiments based on Matlab demonstrate that this method achieves average localisation accuracies of 94.3%, 85.3%, and 81.3%, and recognition accuracies of 92.1%, 82.9%, and 79.7% in sunny, rainy, and foggy scenes, respectively. This demonstrates outstanding performance. However, system performance still requires improvement under conditions of extreme dense fog or heavy rain.