Ermita, Carl Lemuel P.; Viray, Liam Kyle Z.
Enhanced you only look once (YOLOv5) object detection algorithm for traffic flow management - Undergraduate Thesis: (Bachelor of Science in Computer Science) - Pamantasan ng Lungsod ng Maynila, 2025
ABSTRACT: This study enhances the YOLOv5 object detection algorithm for improved vehicle detection in complex, real-world traffic environments. The proposed model integrates semantic segmentation, a car-attention mechanism, and advanced data augmentation to boost accuracy and robustness in scenarios with poor lighting cluttered backgrounds, and varied weather conditions. The model was trained and tested on a diverse dataset of real-time traffic scenes, including parked and moving vehicles. Evaluation shows that the enhanced YOLOv5 significantly outperforms the original YOLOv5 in all performance metrics. On the testing dataset, it achieved a precision of 86.36%, recall of 78.06%, F1 score of 0.08195, and [email protected] of 87.15% compared to the original YOLOv5’s 60.04% precision 57.31% recall, 0.5864 F1 score, and 60.92% [email protected]. These results reflect notable improvements in detection accuracy and the ability to reduce false positives in both normal and adverse conditions. Further comparison with newer versions -----YOLOv8, YOLOv10, and YOLOv11----confirms the enhanced YOLOv5’s superiority in F1 score and [email protected], making it a strong candidate for real-world deployment. The consistent performance across challenging datasets demonstrates the model’s adaptability and efficiency. This research contributes to the development of more intelligent traffic management systems by offering a robust, real-time vehicle detection solution. The findings support its practical application in smart traffic lights, congestion monitoring, and urban mobility systems requiring high-precision detection under varying environmental conditions.
academic writing
QA76.9 A43 E76 2025
Enhanced you only look once (YOLOv5) object detection algorithm for traffic flow management - Undergraduate Thesis: (Bachelor of Science in Computer Science) - Pamantasan ng Lungsod ng Maynila, 2025
ABSTRACT: This study enhances the YOLOv5 object detection algorithm for improved vehicle detection in complex, real-world traffic environments. The proposed model integrates semantic segmentation, a car-attention mechanism, and advanced data augmentation to boost accuracy and robustness in scenarios with poor lighting cluttered backgrounds, and varied weather conditions. The model was trained and tested on a diverse dataset of real-time traffic scenes, including parked and moving vehicles. Evaluation shows that the enhanced YOLOv5 significantly outperforms the original YOLOv5 in all performance metrics. On the testing dataset, it achieved a precision of 86.36%, recall of 78.06%, F1 score of 0.08195, and [email protected] of 87.15% compared to the original YOLOv5’s 60.04% precision 57.31% recall, 0.5864 F1 score, and 60.92% [email protected]. These results reflect notable improvements in detection accuracy and the ability to reduce false positives in both normal and adverse conditions. Further comparison with newer versions -----YOLOv8, YOLOv10, and YOLOv11----confirms the enhanced YOLOv5’s superiority in F1 score and [email protected], making it a strong candidate for real-world deployment. The consistent performance across challenging datasets demonstrates the model’s adaptability and efficiency. This research contributes to the development of more intelligent traffic management systems by offering a robust, real-time vehicle detection solution. The findings support its practical application in smart traffic lights, congestion monitoring, and urban mobility systems requiring high-precision detection under varying environmental conditions.
academic writing
QA76.9 A43 E76 2025