Panoramic Video Stitching in Manufacturing Workshops Based on Key Target Protection

Aiming at the problem that key targets in panoramic surveillance videos of factory manufacturing workshops are split by seams or suffer from ghosting due to parallax, a real-time video stitching method for protecting key targets was studied. Firstly, the homography matrix between the first frames of the videos was estimated. Then, the mask image of the key target was extracted based on the Segment Anything Model, and the mask image of the moving target was extracted by improving the background subtraction method based on morphological filtering. The key target term was constructed based on these two mask images. Finally, the optimal seam line algorithm was improved based on the key target term. The experimental results show that this method can effectively avoid ghosting and cutting of key targets at the seams while ensuring the stitching speed. The obtained stitching results are smooth and natural, outperforming several current stitching methods. This method effectively solves the problem of misalignment and ghosting of key targets in panoramic surveillance videos of manufacturing workshops and performs well in multiple quantitative indicators.