A High-Accuracy Image Stitching Method from a Dual-View Camera System

With the development of computer vision technologies, the view angle of a single-lens camera system has gradually become unsuitable for high-precision remote sensing, surveying and mapping applications. This paper proposes an image stitching method for a dual-view camera system. The method consists of object distance estimation and image stitching steps. Firstly, the object distance was estimated by employing a least squares algorithm based on corresponding feature points in the overlapping area. Then, the Delaunay triangulation method was applied to conduct distance interpolation, which effectively reduces discontinuities in depth transitions. Secondly, the virtual images was generated based on geometric relationship between two cameras and the estimated object distance. Experimental results demonstrate that the proposed method significantly reduces the geometric errors in a stitching image caused by significant depth variations and improves the overall visual consistency.