Vision-Guided Segregation of Biomedical Waste Using AI-Driven Control Architecture

Biomedical waste continues to increase in volume worldwide, particularly in developing regions where improper segregation, inadequate staff training, and inconsistent compliance with color-coded disposal protocols create significant health and environmental risks. Incorrect handling of hazardous biomedical waste enables the transmission of more than 30 high-risk pathogens, including HBV, HCV, and HIV. This growing challenge highlights the need for reliable and automated waste segregation solutions. This paper presents a fully automated biomedical waste segregation device that combines real-time computer vision with precision electromechanical actuation. The system uses a YOLOv8 object detection model trained on a custom dataset of waste items captured under fixed-platform imaging conditions. This approach enables stable classification of common biomedical waste types such as gloves, syringes, vials, and cotton materials. An embedded processor performs both model inference and decision-making, and controlling the motor actuation for the release of the waste into their specific compartments. The system operates on a single-frame inference workflow and is designed to prioritize high precision in order to prevent incorrect bin assignments while maintaining acceptable recall during practical operation. The compact and low-latency architecture provides reliable performance across varied lighting and clinical environments, offering a practical and safe solution for automating waste segregation at the point of disposal. In addition to reducing manual handling and exposure risks for healthcare personnel, the platform can be expanded with IoT-based monitoring for waste tracking, analytics, and regulatory compliance. These capabilities make the proposed system a scalable and cost-effective tool for improving biomedical waste management in resource-constrained healthcare facilities and supporting broader public health and sustainability initiatives.