Yolov5-based rotating target pose grasping
Jian Ying Yang, Shuai Jiang, Kai Chen, Li Liu · 2022
The manufacturing industry is the backbone of the country and supports the development of the national economy. Due to the wide variety of industrial parts and the extreme complexity and variability of the industrial environment, artificially intelligent grasping and sorting technology has raised higher requirements. This has made machine learning-based, computer vision neural networks for visual grasping technology the focus of research in the field of industrial intelligent manufacturing. However, the focus of traditional common computer vision neural networks currently lies mainly on target detection and recognition classification, while the industrial work environment is cluttered with industrial parts placement and the current grasping pose problem has to be implemented with the help of non-neural network methods. This leads to a bloated and complex grasping and sorting system and is highly prone to problems such as missed grasping and misgrasping under problematic conditions such as mutual occlusion, which hinders the application and development of intelligent industrial production. In this paper, from the perspective of improving the algorithmic function of deep neural networks, we propose a target recognition and pose localization algorithm based on the Yolov5 model to achieve continuous and effective classification recognition localization and pose grasping of industrial parts moving on a conveyor belt using only one neural network.