Research on cylindrical labels unfolding algorithm based on machine vision

It is a difficult problem for imaging the cylindrical label, and this paper presents an algorithm for unfolding the cylindrical label. First, the cylindrical label is positioned by the monocular camera, and the 3D point sets of the cylindrical label are established. Second, the observation range of each camera is determined based on the poses of the cylinders. Then, the cylindrical distortion is corrected in each camera. Finally, two adjacent images are stitched together to obtain a continuous and normalized unfolding image of the cylindrical label. Real-world and synthetic data experiments results show that the algorithm can correct the cylindrical distortion, and it takes only about 160 ms to obtain the cylindrical label unfolding image. The algorithm execution process does not require complicated steps and additional auxiliary, and the algorithm execution speed meets the industrial requirements.