| dc.description.abstract | As the world population becomes larger, the number of vehicles on the road increases
dramatically, leading to a rise in road violations. One of the most common types of violation is
red-light running (RLR). This thesis presents a comprehensive study about a cost-effective RLR
detection system. By incorporating neural networks and object detection methods, the
development of this system was able to detect RLR violations with high accuracy. This thesis
presents a comprehensive study of a new approach for red light running detection system using
the YOLOv8 detection model. The main goal of this research is to improve road safety by
developing a precise and dependable method of identifying traffic signal offenses, with a focus
on red-light violations. This thesis uses state-of-the-art YOLO architecture to develop a reliable
and affordable solution that can be easily applied to stop the increasing number of traffic
violations related to driving under red lights. The research involves various processes, such as
collecting datasets from CCTV footage and different sources across the internet. Additionally,
different techniques, such as data augmentation, are applied to enhance the model's
performance in real-world applications. The processes include training, validating, and testing
the object detection model. To track violations, the research also incorporates DeepSORT as
the proposed tracking algorithm on top of YOLOv8 to flag detected violators. Furthermore,
there is an implementation of a graphics user interface (GUI) to facilitate easy navigation and
handling for potential users. The end system was able to achieve relatively high performance,
with 92.4% precision, 91.2% recall, and 94.0% mAP50. | en_US |
