Artificial intelligence and machine learning

Applying the YOLOv11 model and the adapted LaDD dataset to people detection in sparsely populated areas

A. V. Smirnov, I. P. Tishchenko, S. A. Lazarev

Ailamazyan Program Systems Institute of RAS, Ves'kovo, Russia

Abstract: This study is aimed at developing a neural network method for detecting people in sparsely populated areas using images obtained from an unmanned aerial vehicle (UAV). The YOLOv11m architecture model was used as a neural network detector. As part of the study, an adaptation algorithm for the LaDD training dataset was developed and applied. Experiments were conducted to preliminary train the model on the original and adapted datasets, which demonstrated the advisability of using the adapted dataset. The final accuracy of the model during training reached $98.7\%$ by metric $mAP^{50}$. Model inference showed a detection accuracy of $0.895$ ($89.5\%$) by metric $\mathrm{F1}$ and $0.901$ ($90.1\%$) by metric $mAP^{50}$, which confirms the workability of the presented method.

Key words and phrases: Image analysis, people detection, UAV imagery, YOLOv11, neural networks, dataset, adaptation.

UDC: 004.932.72
BBK: 32.813.52

MSC: Primary 68T10; Secondary 68T45

Received: 28.10.2025
Accepted: 09.11.2025

DOI: 10.25209/2079-3316-2025-16-4-217-240