Tools
Tools
Shrivastava, Lucky (2023) Real-Time Pedestrian Detection using YOLO-NAS. Masters thesis, Dublin, National College of Ireland.
Preview |
PDF (Master of Science)
Download (2MB) | Preview |
Preview |
PDF (Configuration Manual)
Download (972kB) | Preview |
Abstract
This research delves into the dynamic field of object detection, particularly focusing on the crucial realm of pedestrian detection. Drawing upon advancements in machine learning and deep learning, this study explores the effectiveness of the YOLO-NAS technique—an efficient real-time detection method based on Single Stage Detection (SSD). The importance of pedestrian detection is underscored by the rising incidents of accidents and fatalities. The YOLO-NAS technique, a derivative of the YOLO framework, has proven successful across diverse applications, including object detection, security, monitoring, and safety. This research employs YOLO-NAS for pedestrian detection using a Kaggle-acquired dataset consisting of video sequences transformed into 640x640 pixel images with applied bounding boxes for accurate prediction. Four models, labeled Model 1 to Model 4, were trained for varying epochs. The third model achieved an outstanding mean Average Precision (mAP) of 0.67, outperforming the other models. Visualizations further enhance the understanding of the obtained result. The trained model can be further honed to improve its accuracy and can be employed in real-time environment for detecting pedestrians.
Actions (login required)
 |
View Item |