| 标 题： |
A Dynamic Weight Adjustment System for Aerial Video Object Recognition Using YOLOv8 with Night Vision Integration
基于YOLOv8的空中视频目标识别动态重量调整系统，集成夜视功能 |
| 作 者： |
Federico Candela
Claudio De Capua
Andrea Morabito
Francesco Carlo Morabito
Francesco Restuccia |
| 关键词： |
Dynamic weight adjustment [动态重量调节]
YOLOv8 [ YOLOv8]
VisDrone [ VisDrone]
Simulated night vision [ 模拟夜视] |
| 出版年： |
2024 |
| 文集名： |
2024 IEEE International Workshop on Technologies for Defense and Security (TechDefense 2024)
2024年IEEE国际国防与安全技术研讨会（TechDefense 2024） |
| 会议名： |
2024 IEEE International Workshop on Technologies for Defense and Security (TechDefense 2024)
2024年IEEE国际国防与安全技术研讨会（TechDefense 2024） |
| 届 次： |
|
| 摘 要： |
This article presents an innovative system for object recognition in aerial and ground-based video, leveraging YOLOv8 with the integration of night vision capabilities. The system addresses the challenges of object detection in dynamic environmental conditions by adapting model weights in realtime based on variations in drone altitude and distance. The proposed methodology ensures high detection performance even in complex scenarios, such as changes in lighting or perspective. In experimental tests conducted on aerial and ground video representing a variety of operational scenarios, the dynamic system achieved an average precision of 89.93% and an average recall of 90.70%, significantly outperforming traditional models like YOLOv8 and VisDrone. The integration of night vision, implemented through the simulation of an image intensifier, further enhanced visibility in low-light conditions. These results demonstrate the effectiveness of the proposed dynamic system in providing a robust solution for real-time monitoring in variable operational environments.
本文提出了一种创新的空中与地面视频目标识别系统，结合YOLOv8模型并集成夜视功能。该系统通过根据无人机高度和距离的变化实时调整模型权重，有效应对动态环境下的目标检测挑战。所提出的方案即使在光照或视角变化等复杂场景下，仍能保持高检测性能。在针对多种实际操作场景的空中与地面视频进行的实验测试中，该动态系统实现了平均准确率89.93%和平均召回率90.70%，显著优于YOLOv8和VisDrone等传统模型。通过模拟图像增强器实现的夜视功能进一步提升了低光条件下的可见性。这些结果表明，所提出的动态系统在提供适用于多变运行环境的实时监控稳健解决方案方面具有显著有效性。 |
| 中文内容为网络AI翻译，专业术语未必准确，仅供参考。 |
京公网安备 11010202009788