Hybrid Approach to Pedestrian Detection: Integrating Transfer learning and Training from Scratch

Goyal, Abhishek (2024) Hybrid Approach to Pedestrian Detection: Integrating Transfer learning and Training from Scratch. Masters thesis, Dublin, National College of Ireland.

Preview	PDF (Master of Science) Download (1MB) | Preview
Preview	PDF (Configuration Manual) Download (1MB) | Preview

Abstract

In current research work, the enhancements in pedestrian detection have been investigated using YOLOv5 and YOLOv8 deep learning models based on hybrid training. The study compares three strategies: There are three categories they discussed, namely, transfer learning, in which a pre-trained model is used while discarding all its layer and training new layers from scratch; the training from scratch technique where a new model is trained completely; and the hybrid method, where new layers while some layers of the pre-trained model are fine-tuned to accomplish the preferred task. All the models were trained using a custom pedestrian dataset, and the performances were assessed using measures of precision, recall, mean average precision (mAP), and inference time. The experiments indicate that the combination approach is superior to the two benchmark methods, with YOLOv8 hybrid yielding the highest accuracy and recall rates and YOLOv5 hybrid being the fastest in inference time. The hybrid models also exhibited excellent performance robustness in real world conditions like occlusion, scale variability, light variations and the like. Such results indicate that the hybrid training approaches, where some layers are frozen and others are fine-tuned, allow achieving both high precision and reasonable running time. The findings of this study indicate that hybrid model Real-time pedestrian detection can be effectively implemented in autonomous driving, security and smart city science. It also exposes directions for future research on enhancing and improving other forms of hybrid models for other object detection problems and the practical implementation of such models in environments with limited resources.

Item Type:	Thesis (Masters)
Supervisors:	Name Email Del Rosal, Victor UNSPECIFIED
Subjects:	Q Science > QA Mathematics > Electronic computers. Computer science T Technology > T Technology (General) > Information Technology > Electronic computers. Computer science T Technology > TL Motor vehicles. Aeronautics. Astronautics Q Science > QH Natural history > QH301 Biology > Methods of research. Technique. Experimental biology > Data processing. Bioinformatics > Artificial intelligence > Computer vision Q Science > Q Science (General) > Self-organizing systems. Conscious automata > Artificial intelligence > Computer vision
Divisions:	School of Computing > Master of Science in Artificial Intelligence
Depositing User:	Ciara O'Brien
Date Deposited:	20 Jun 2025 08:21
Last Modified:	20 Jun 2025 08:21
URI:	https://norma.ncirl.ie/id/eprint/7951

Actions (login required)

View Item