A Novel Approach to Generate Dataset for Object Detection in Assembly Lines

Quality Assurance (QA) is required to ensure precision in the vehicle Assembly Unit process. The most significant challenge is that manual work is error-prone, and even minor errors can be a problem for a vehicle. This study aims to explore suitable Deep Learning (DL) models to automate various parts of the task well. For the current work, the aim is focused on predicting/detecting points on the chassis accurately. In this article, we elaborated on the process to generate the dataset and the proposed model 'You Look Only Once-v5' (YOLOv5) to identify cross marks on the vehicles. The model architecture and parameters are discussed in-depth and changed to detect and classify marked objects against the chassis background. The accuracy and efficiency evaluations show that the model achieved the top performance in average precision (mAP) of ≥ 98%.