Optimized YOLOv8 Model for Aerial Pedestrian Detection in Drone-Based Monitoring Systems

Zijun Wang, Huimin Meng, Junjie Liu, Ge Meng
Proceedings of 2025 2nd International Conference on Machine Learning and Intelligent Computing, PMLR 278:395-403, 2025.

Abstract

The widespread application of unmanned aerial vehicles (UAVs) in emergency rescue and security inspection poses stringent demands for accuracy and real-time performance in small-target personnel detection from aerial perspectives. Addressing the limitations of existing algorithms in complex background interference, multi-scale targets, and feature sparsity, this paper proposes an improved lightweight YOLOv8 detection model. By designing a multi-dimensional attention collaboration module to enhance feature focus, constructing a high-resolution detection layer to improve shallow feature utilization, and optimizing localization accuracy with geometrically constrained loss functions, the method achieves an 11.68% detection accuracy improvement over the baseline model on UAV datasets while maintaining real-time processing at 158 FPS. It effectively resolves small-target missed and false detection issues, providing reliable technical support for UAV-based intelligent inspection tasks.

Cite this Paper

BibTeX
@InProceedings{pmlr-v278-wang25f, title = {Optimized YOLOv8 Model for Aerial Pedestrian Detection in Drone-Based Monitoring Systems}, author = {Wang, Zijun and Meng, Huimin and Liu, Junjie and Meng, Ge}, booktitle = {Proceedings of 2025 2nd International Conference on Machine Learning and Intelligent Computing}, pages = {395--403}, year = {2025}, editor = {Zeng, Nianyin and Pachori, Ram Bilas and Wang, Dongshu}, volume = {278}, series = {Proceedings of Machine Learning Research}, month = {25--27 Apr}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v278/main/assets/wang25f/wang25f.pdf}, url = {https://proceedings.mlr.press/v278/wang25f.html}, abstract = {The widespread application of unmanned aerial vehicles (UAVs) in emergency rescue and security inspection poses stringent demands for accuracy and real-time performance in small-target personnel detection from aerial perspectives. Addressing the limitations of existing algorithms in complex background interference, multi-scale targets, and feature sparsity, this paper proposes an improved lightweight YOLOv8 detection model. By designing a multi-dimensional attention collaboration module to enhance feature focus, constructing a high-resolution detection layer to improve shallow feature utilization, and optimizing localization accuracy with geometrically constrained loss functions, the method achieves an 11.68% detection accuracy improvement over the baseline model on UAV datasets while maintaining real-time processing at 158 FPS. It effectively resolves small-target missed and false detection issues, providing reliable technical support for UAV-based intelligent inspection tasks.} }
Endnote
%0 Conference Paper %T Optimized YOLOv8 Model for Aerial Pedestrian Detection in Drone-Based Monitoring Systems %A Zijun Wang %A Huimin Meng %A Junjie Liu %A Ge Meng %B Proceedings of 2025 2nd International Conference on Machine Learning and Intelligent Computing %C Proceedings of Machine Learning Research %D 2025 %E Nianyin Zeng %E Ram Bilas Pachori %E Dongshu Wang %F pmlr-v278-wang25f %I PMLR %P 395--403 %U https://proceedings.mlr.press/v278/wang25f.html %V 278 %X The widespread application of unmanned aerial vehicles (UAVs) in emergency rescue and security inspection poses stringent demands for accuracy and real-time performance in small-target personnel detection from aerial perspectives. Addressing the limitations of existing algorithms in complex background interference, multi-scale targets, and feature sparsity, this paper proposes an improved lightweight YOLOv8 detection model. By designing a multi-dimensional attention collaboration module to enhance feature focus, constructing a high-resolution detection layer to improve shallow feature utilization, and optimizing localization accuracy with geometrically constrained loss functions, the method achieves an 11.68% detection accuracy improvement over the baseline model on UAV datasets while maintaining real-time processing at 158 FPS. It effectively resolves small-target missed and false detection issues, providing reliable technical support for UAV-based intelligent inspection tasks.
APA
Wang, Z., Meng, H., Liu, J. & Meng, G.. (2025). Optimized YOLOv8 Model for Aerial Pedestrian Detection in Drone-Based Monitoring Systems. Proceedings of 2025 2nd International Conference on Machine Learning and Intelligent Computing, in Proceedings of Machine Learning Research 278:395-403 Available from https://proceedings.mlr.press/v278/wang25f.html.

Related Material