The research on an image stitching scheme for remote driving scenarios of vehicles

Abstract

As remote driving technology develops, higher requirements are imposed on the environmental perception capabilities during remote driving. An image stitching scheme for remote driving scenarios of vehicles is proposed to reduce the visual blind spots of remote driving vehicles and enhance environmental perception capabilities during remote driving. The images to be stitched are processed using distortion correction, cylindrical projection, and image registration in preparation for subsequent stitching. The energy function in the stitching process is improved to reduce errors in finding the optimal seam, resulting in a higher quality seam than traditional energy functions. At the same time, finding the optimal seam speed is increased by 257.0% on average by introducing the optimization process of coarse sampling, restoration, and fine sampling into the dynamic programming algorithm to find the optimal seam, compared with the direct finding optimal seam in the images overlap area. In addition, since the traditional image stitching algorithm is incapable of completely eliminating the ghosting phenomenon, an algorithm that eliminates seam line traces based on the brightness difference at the optimal seam is innovatively adopted in this scheme, due to the absence of image fusion, the generation of ghosting is fundamentally eliminated. Finally, experiments with remote driving vehicles are conducted to stitch environmental images encountered in remote driving scenarios, the experimental results indicate that the proposed scheme is well-suited for image stitching in vehicles’ remote driving scenarios, ghosting is effectively eliminated, stitching transitions smooth and natural color balance are ensured, and the overall stitching image quality is the best compared to traditional algorithms.