Automated label quality control using computer vision and a collaborative robot

Student: Dario Levanic

This work presents the development of an automated inspection system for detecting defects on bottle labels and removing faulty products from a conveyor line using computer vision and a collaborative robot.

The system integrates multiple components, including an ultrasonic sensor, conveyor control, image acquisition, deep-learning-based defect detection, and robotic manipulation. Bottle presence is detected using an ultrasonic sensor connected to an Arduino microcontroller, which controls the conveyor belt. Once a bottle is positioned, the system captures an image of the label using a mobile device configured as an IP camera, with image processing handled via OpenCV.

Defect detection is performed using a YOLO-based object detection model, initially pretrained on the COCO dataset and further fine-tuned on a custom dataset containing examples of wrinkled labels, peeled edges, and correct labels. Based on the detection results, the system classifies bottles as acceptable or defective. In the case of detected defects, a rejection routine is triggered via the Universal Robots UR Dashboard interface, enabling automatic removal of faulty items from the production line.

This work demonstrates a practical integration of AI, computer vision, and robotics for quality control in industrial environments, with potential for further improvements and scalability in smart manufacturing systems.

Repository: https://repozitorij.fsb.unizg.hr/object/fsb:13678