Assembly progress estimation using instance segmentation with data augmentation considering annotation data

In this study, we propose a method for estimating product assembly progress in a factory using instance segmentation. The method detects parts by instance segmentation and estimates the progress based on the number and types of detected parts. The background affects the accuracy of estimation, and the training data is limited. Therefore, we apply data augmentation using random masking to reduce the effect of background information and improve part detection accuracy. Experiments are conducted on a shelf simulating a factory product to confirm the effectiveness of the proposed method.