Research on robotic monocular 6DOF vision localization grasping strategy for industrial objects

PurposeTo address the issue of large visual measurement errors caused by insufficient information collected by monocular vision when performing six-degree-of-freedom (6DOF) position measurements on metal castings, which hinders the robot's ability to visually guide grasping, this paper aims to propose a 6DOF position measurement method that integrates monocular vision with deep neural networks.Design/methodology/approachThis method enhances the robot's ability to visually grasp small-sample industrial objects with high accuracy. By establishing a mapping relationship between the two-dimensional (2D) position of the object's image and its three-dimensional (3D) position in space, the proposed approach achieves 6DOF position measurement of the target workpiece using monocular vision. An image enhancement algorithm based on a generative adversarial network (GAN) is introduced to improve robustness in industrial environments by addressing the challenge of acquiring image data for small-sample objects. Additionally, the method combines single-phase object detection using deep neural networks with 2D-3D affine transformation to achieve accurate 3D position measurements.FindingsThe introduction of the GAN-based image enhancement algorithm significantly mitigates the robustness issues posed by the difficulties in obtaining image data for small-sample objects in industrial settings. The integration of single-phase object detection and 2D-3D affine transformation allows for precise 3D position measurement of the workpiece. Experimental results demonstrate that the proposed method provides high accuracy in 6DOF position measurements for industrial objects.Originality/valueThis approach overcomes the limitations of traditional vision algorithms for 3D position measurement of industrial objects, such as high cost and poor robustness. The experimental validation confirms that the proposed method achieves excellent 6DOF position measurement accuracy for industrial objects.