Anchor-free Hand-detection with Lightweight Design for Smart Factories

Abstract

Nowadays, good hand detection has been proven helpful for a smart assembly factory to improve work efficiency. Particularly, an assembly line for an Industry 4.0 factory needs to manufacture a diverse range of products, its assemblers must acquire knowledge of distinct assembly processes and inspection procedures. Therefore, hand detection via cameras has become a prevalent method of aiding the assembly process in smart factories. However, existing hand detection in a smart factory still relies on a powerful back-end server for image processing due to the limited computing power of a camera. To address this issue, we propose a “Lightweight Anchor-Free Hand-detection Model” (LAFHDM), and the resultant deep neural networks (DNNs) can directly fit into a smart camera to detect assemblers’ hand positions to verify the correctness of assembly steps. The proposal is also in accordance with the inevitable trends of edge computing and the Internet of Things. The experimental results show that the inference speed of the LAFHDM is at least 40 times faster, the accuracy can be 27.41% higher than that of previous models. Moreover, the inference speed of an edge camera is improved approximately three times.