Intelligent splicing method for multi-view-field images of integrated cabinet door of intelligent machine room
By employing deep learning and global optimization techniques, the challenge of stitching weak texture images of server rack doors in intelligent data centers has been solved, achieving high-precision, seamless panoramic image generation and improving stitching quality and consistency.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- 国网宁夏电力有限公司信息通信公司
- Filing Date
- 2026-02-05
- Publication Date
- 2026-05-15
AI Technical Summary
Existing technologies struggle to perform high-quality image stitching on cabinet doors in smart data centers, especially in areas with weak textures where feature matching rates are low and mismatch rates are high. Furthermore, traditional methods are prone to producing stitching misalignments, ghosting, and brightness/color seams, and lack an effective global error optimization mechanism.
By employing deep integration of unsupervised deep learning, global optimization theory, and advanced image fusion technology, multi-view images are acquired through a collaborative acquisition system. Image registration is performed using a pre-trained unsupervised deep image alignment network. A global binding adjustment and optimization model is constructed, and seamless panoramic images are generated using the best stitching seam search algorithm and multi-band fusion technology.
It achieves high-precision, fully automatic, and high-fidelity stitching of weak-texture cabinet door images, significantly improving the feature matching success rate, eliminating accumulated errors, and generating seamless panoramic images with excellent visual quality.
Smart Images
Figure CN122048645A_ABST