Leveraging Foundation Model Automatic Data Augmentation Strategies and Skeletal Points for Hands Action Recognition in Industrial Assembly Lines

arXiv:2403.09056v1 Announce Type: new
Abstract: On modern industrial assembly lines, many intelligent algorithms have been developed to replace or supervise workers. However, we found that there were bottlenecks in both training datasets and real-time performance when deploying algorithms on actual assembly line. Therefore, we developed a promising strategy for expanding industrial datasets, which utilized large models with strong generalization abilities to achieve efficient, high-quality, and large-scale dataset expansion, solving the problem of insufficient and low-quality industrial datasets. We also applied …

abstract action recognition algorithms arxiv assembly augmentation bottlenecks cs.cv data datasets found foundation foundation model however industrial intelligent line modern performance real-time recognition strategies training training datasets type workers