Optimizing Traffic Law Enforcement: A Deep Learning Approach to Philippine Number Coding

This study addresses the persistent issue of traffic congestion in Metro Manila, particularly focusing on the effectiveness of traffic management programs, such as the Modified Unified Vehicular Volume Reduction Program (UVVRP). This study investigates the use of Faster R-CNN and YOLOv7 for license plate detection and coding violation identification, aiming to improve the efficiency of automated enforcement systems like the No-Contact Apprehension Policy (NCAP). The dataset, collected from busy streets along Boni Avenue, Mandaluyong, includes real-world traffic scenarios and emphasizes the identification of number coding violators. The study evaluates the models' performance using metrics such as mean average precision (mAP) and explores the impact of including DeepSORT in the YOLOv7 model. Results indicate Faster R-CNN's superiority in detecting license plates and coding violations with a mean Average Precision of 98% and 88% at 0.50 IoU, respectively, with YOLOv7 demonstrating competence. In conclusion, Faster R-CNN excels in accuracy, while YOLOv7 with DeepSORT enhances real-time detection, offering insights for traffic regulation improvements in Metro Manila cities.