Three-Step Strategy for Pattern Recognition and Rotation Angle Estimation of Rectangular Workpieces

The burgeoning field of automated assembly is undergoing rapid evolution, thanks to the recent strides in deep learning and computer vision technologies. However, the journey is marred by significant challenges, particularly inaccurate classification precision and suboptimal positioning accuracy, which stifles technological progression. To address these challenges, this study proposes a new Swin Transformer and ORB (STO) algorithm, aimed at improving the classification, positioning, and rotation accuracy of key components, especially rectangular objects, in automated assembly lines. The STO algorithm consists of three main components: a Swin Transformer-based object classification system, a positioning model for rectangular objects, and a model for calculating rotation angles. The positioning model uses the techniques of threshold processing and contour detection to locate rectangular objects effectively. Meanwhile, the rotation angle calculation model employs the oriented FAST and rotated BRIEF(ORB) algorithm for feature extraction and matching, ensuring precise determination of the required rotation angles. This study sets up an experimental apparatus including a camera, a robotic arm, and randomly placed rectangular workpieces. The randomly placed rectangular workpieces are regarded as rectangular workpieces that need to be assembled. Results demonstrate that the STO algorithm excels in object recognition and angle determination, particularly showing high precision in angle calculation, with a mean absolute error (MAE) of 0.10°. In summary, the proposed method improves the accuracies of rotation angle estimation and pattern recognition of the workpieces, thereby showing potential applications in the industrial assembly process. The STO method principle also shows potentials in recognizing irregular workpieces in various industrial scenarios.