Machine Vision for Sorting

Abstract

The integration of robots and machine vision in manufacturing has significantly advanced automation, reducing labour costs and improving productivity and safety. This thesis investigates the effectiveness of three object detection models—YOLO, SSD, and Faster R-CNN—in classifying and localizing mechanical fasteners for industrial sorting. The models were trained and evaluated by integrating with Nyrio Ned2 vision set. The performance was measured in terms of accuracy, precision, recall, F1 score, false negative rate, and inference time. YOLO exhibited the highest accuracy and a balanced performance, while Faster R-CNN achieved the highest precision but lagged in inference speed. SSD, although the fastest, showed the poorest accuracy and recall. Real-time tests under varying scenarios further validated the models' robustness and adaptability. The results suggest that YOLO provides a suitable balance between precision, recall and speed for real-time sorting of mechanical fasteners.