An ordinary traffic surveillance camera has difficulty capturing all the information due to its limited field of view. This paper proposes a method for stitching traffic images with geometric feature protection (GFP) to expand the view and keep the panorama as natural as possible. The core of this method is to build up a two-stage mesh optimization model integrating shape preservation, line protection, and boundary protection. In terms of boundary protection, a boundary regularization term is added to protect the boundaries of the initial stitched images to further achieve a regular display effect. The experiments show that the proposed method can handle traffic images well. It further enhances the alignment of lines and regularizes image boundaries whereas maintaining a high degree of similarity, making panoramas more natural and visually acceptable than several representative methods.
Road Traffic Image Stitching with Geometric Feature Protection
2022-10-08
3077141 byte
Conference paper
Electronic Resource
English