Automated waste sorting has become increasingly important in improving recycling efficiency, reducing human exposure to hazardous materials, and supporting sustainable waste management practices. This study presents the design and implementation of a machine vision-based conveyor sorting system capable of real-time multi-object detection under variable lighting conditions. The proposed system integrates a conveyor mechanism, an RGB imaging sensor, adjustable illumination units, an embedded processing platform, and electromechanical actuators to enable automated detection, classification, and physical separation of selected waste materials. A deep learning–based object detection framework is employed to identify multiple waste objects simultaneously as they move along the conveyor belt. To evaluate robustness against lighting variability, system performance is tested under uniform, low-intensity, and mixed lighting conditions. Key performance metrics include detection accuracy, inference latency, frame processing rate, and overall sorting efficiency. Experimental evaluation of the bench-scale prototype demonstrates reliable real-time performance using costeffective embedded hardware. Results indicate that while detection accuracy is highest under uniform illumination, the system maintains acceptable performance under dim and uneven lighting conditions, with moderate degradation primarily due to shadows and glare. The proposed approach provides a scalable and low-cost solution for intelligent waste sorting applications and establishes a practical foundation for future research, including advanced lighting compensation, depth sensing, and industrial-scale deployment.