An Object Detection Approach for Automated Detection of Groove Line in Tube Yoke

Abstract

Deep Neural Networks are designed explicitly for visionbased applications. They have become a standard method for image classification, object detection, real-time processing capabilities, and other computer vision tasks. With the intervention of human beings, it is challenging for businesses to verify that the product is appropriately created without any glitches, as this could result in human errors due to a lack of training, which could further cause losses and several complications. Thus, there is a need for a machine learning-based solution that improves work efficiency and accuracy in identifying processed and unprocessed parts. Machine Learning models have several advantages over human intervention in tasks, including consistency, speed, scalability, costeffectiveness, and improvement over time. These advantages can help organizations improve their operations and achieve better outcomes. To do this, we propose an approach that involves building a model using an object detection method, You Only Look Once (YOLO), which is then deployed on a Raspberry Pi and integrated with the assembly line to carry out the task of validating the processed parts by achieving 98% accuracy.

Application of YOLOv8 and Detectron2 for Bullet Hole Detection and Score Calculation from Shooting Cards

“…YOLOv8 has five further variants ranging from nano-scale to extra-large models. The choice of these models is to be made based on the tradeoff between the required accuracy and the inference time [26]. In the case of finding the points scored on a shooting card, speed is of significant importance.…”