The Panoramic Video Stitching Method for Dynamic Scenes

To address the problems of significant image distortion, edge aliasing, and unstable dynamic imaging that exist after multi-camera video stitching, this paper proposes a panoramic video stitching method for dynamic scenes. First, the cylindrical projection method is improved using backward mapping and bilinear interpolation. Then, spatial detection constraints are set in the feature detection procedure to enhance processing efficiency and accuracy. In the matching procedure, initial image matching is performed through ratio testing, followed by a dual matching mechanism that includes bidirectional cross-validation of coarse matching and RANSAC for fine matching, thereby further eliminating mismatched points and improving the quality and stability of video fusion. Finally, the feasibility and effectiveness of the proposed method are verified both through simulations and experiments. The proposed method can be extended to fields such as smart transportation, intelligent manufacturing, security monitoring, and remote collaboration, providing real-time and efficient wide-area visual perception solutions for Industrial Internet of Things and smart cities.