Fine-tuned YOLO Model for Monitoring Children Across Medical Scenes Based on a Large-Scale Real-World Dataset for Children Detection

Accurate detection and monitoring of children in medical settings using computer vision systems present unique challenges due to anatomical differences, environmental complexity, and stringent privacy constraints. This paper introduces YOLOCDD , a fine-tuned YOLOv11-based model optimized for child detection in medical scenes, supported by the Child Detection Dataset (CDD)—a large-scale, real-world dataset comprising 1,928 annotated images of children across diverse age groups and interaction scenarios. Unlike existing datasets that rely heavily on synthetic data or controlled environments, CDD captures realistic medical and everyday settings, including occlusions, multi-child interactions, and dynamic lighting conditions. Our model achieves a mean average precision (mAP@50) of 0.953 in medical environments, significantly outperforming general-purpose detectors like YOLOv11x (mAP@50: 0.606).This work bridges critical gaps in pediatric medical AI by providing a scalable, privacy-compliant dataset, delivering a high-precision detection model, and showcasing clinical applicability in neurological diagnostics. The dataset is publicly available to foster further research in child-centric computer vision.