Object Detection and Tracking, Searching Using Computer Vision and YOLO Technique

Abstract

Real-time visual understanding has become a cornerstone of modern intelligent systems, spanning surveillance, autonomous navigation, retail analytics, and smart-city infrastructure. The system architecture is described in detail, including data-flow diagrams, use-case and sequence diagrams, an activity diagram, and a relational database design for persisting detection and tracking metadata. An experimental evaluation performed on standard benchmark-style data demonstrates that the proposed pipeline attains a mean Average Precision (mAP@0.5) of approximately 0.91, a tracking identity-switch rate reduced by 38% relative to a naive frame-by-frame detector, and a sustained throughput of 61 frames per second on a mid-range GPU, outperforming several baseline architectures compared in this study. The results confirm that combining YOLOv8 with Deep SORT and a dedicated search layer yields a practical, extensible platform for real-time object detection, tracking, and retrieval applications.