Computer Vision Intelligence for Medication Safety: Pill Instance Segmentation in Philippine Healthcare

Medication errors arising from pill misidentification remain a significant concern in healthcare, particularly in the Philippines where generic and branded medicines frequently share similar visual characteristics. This study introduces a computer vision-based system for automated identification and classification of locally distributed pharmaceutical pills using the YOLOv12 instance segmentation model. A localized dataset consisting of 20 Philippine medicine classes and approximately 6,000 high-resolution images was curated and annotated using Roboflow's smart segmentation tools. The proposed model achieved a segmentation mAP@0.50:0.95 of 0.935 and a near-perfect mAP@0.50 of 0.980, demonstrating strong accuracy, generalization capability, and robustness under varied lighting conditions and partial occlusions. These results outperform several existing international pill-recognition systems while addressing regionspecific challenges in the Philippine pharmaceutical landscape. The model's efficiency, precision, and deployment readiness highlight its potential for real-world use in medication verification, supporting patient safety initiatives through accessible, AI-driven, mobile-compatible healthcare solutions tailored to low-resource environments.