Real-Time Multiple Object Detection and Tracking Using YOLOv8: An Interactive Deep Learning Framework with ByteTrack and BoT-SORT

This technical report presents the design, implementation, and evaluation of a real-time multiple object detection and tracking system using the Ultralytics YOLOv8 deep learning framework. The system combines YOLOv8-based object detection with ByteTrack and BoT-SORT tracking capabilities to support persistent multi-object tracking across video frames.

A custom YOLOv8n model was trained using an annotated horse-racing dataset, with images processed at a resolution of 640 × 640 pixels. The implementation integrates Python, PyTorch, OpenCV, Roboflow, and Streamlit to provide an interactive application capable of processing images, prerecorded videos, and webcam streams.

The original experimental study reported overall performance of 73.8% precision, 81.7% recall, and 76.8% mean Average Precision (mAP). A separate horse-detection experiment reported 87.83% precision, 91.86% recall, and 93.84% mAP. The system demonstrates the practical application of deep learning-based computer vision to real-time object detection and tracking, including scenarios involving multiple and partially occluded objects.

This technical report develops and extends experimental work originally undertaken as part of the author's MSc Artificial Intelligence dissertation at Sheffield Hallam University. It provides a refined account of the system architecture, model development, tracking implementation, experimental findings, limitations, and opportunities for future research.