SINGLE SHOT DETECTOR-EFFICIENTDET (SSD-ED) MODEL FOR REAL-TIME MALAYSIAN NUMBER PLATE DETECTION AND RECOGNITION

Abstract

Real-time number plate (NP) detection and recognition play a crucial role in intelligent transportation systems, enabling automated toll collection, smart parking systems, and traffic management. Despite advances in deep learning (DL) frameworks, various challenges persist in achieving robust performance across different scenarios, such as variations in languages, font types, colors, formatting regulations, illumination level, occlusion, display angle, and weather. This paper aims to propose a reliable and efficient DL framework model for real-time Malaysian NP detection and recognition. The single-shot detector-efficientdet (SSD-ED) DL-based model and the YouTube Malaysian NP dataset are used for NP detection and recognition. All SSD-ED variants are evaluated for their performance in NP detection and alphanumerical character recognition tasks. The results demonstrate the superiority of SSD-ED7 in accuracy, with 94.6% for NP detection and 93.4% for character recognition. However, it has longer processing times than other variants. SSD-ED4, on the other hand, shows balanced performance and speed, with accuracy rates of 91.6% and 90.6% for NP detection and character recognition, respectively. The higher the accuracy, the longer the processing time, making it less suitable for fast-response applications and more appropriate for accuracy-driven ones. Therefore, the SSD-ED4 model is well-suited for real-time applications, providing efficient and accurate NP detection and character recognition.