Automatic Number Plate Recognition Techniques Performance on Zimbabwean Number Plates

In this modern age, Automatic Number Plate Recognition systems are an important aspect to improve service efficiency (toll collection, traffic management, parking management), security (vehicle tracking, monitoring vehicle visits) and law enforcement when dealing with motor vehicles. However, many of the existing Automatic Number Plate Recognition systems were developed for countries with number plate standards and font that differ from the Zimbabwean number plates. This research compares different approaches to Automatic Number Plate Recognition techniques with a special focus on Zimbabwean number plates through experiments. The experiments were grouped into two main phases of Automatic Number Plate Recognition namely plate detection and plate recognition. The experiments for plate detection compared traditional image processing techniques using OpenCV and deep learning techniques using the YOLOv8 algorithm. The YOLOv8 algorithm proved to be the most accurate approach for plate detection. Experiments for plate recognition compared Optical Character Recognition engines and machine learning approaches. The compared machine learning algorithms were random forest classifier, and K-Nearest Neighbour. In this set of experiments, a K-Nearest Neighbour algorithm proved to be quicker and more accurate in detecting Zimbabwean number plates. This research recommends the development of an endto-end system for Zimbabwean number plates recognition using deep learning.