Research On Panorama Stitching Algorithms For Railway Environment Based On Forward Motion Videos

Image stitching technology is an attractive topic of image processing, computer vision and computer graphics research fields. It can be used to create an image that has high-resolution, large viewing angle. Image stitching technology has been widely used in many research fields, such as remote sensing, virtual reality, medical image processing, military, photogrammetry, etc.In recent years, high-speed railway is growing rapidly. With the upgrading of train speed, any little accident could lead to a big consequence. Thus, it has become an extremely urgent issue to ensure the safety of high speed trains. Currently, the typical solution is to use diagnostic train, where the camera was mounted. The train will run over the entire line every day, and the camera will record the entire line during the train running. In the daily patrol, record entire line will take a lot of storage space. Compared with video, panorama has many advantages, such as smaller storage space requirement, easily for quick browsing.This paper proposed some solutions for these problems. The main works of this paper are:1. Based on priori information of forward motion video of trains, we propose a framework for panorama acquisition. This framework divided forward motion videos into four planes, which are up, down, left and right, respectively. Each plane generate a panorama.2. For the forward motion videos which have a poor quality, we propose a fast image stitching method. This method can handle the poor quality videos, which other methods cannot.3. For the forward motion videos which have a good quality, we propose an image stitching method based on adaptive manifolds. This method can deal with the videos which have a complex background.4. Based on the panoramas we created before, we reconstruct these panoramas in3D space, to simulate the real environment of train running. This work can make a quick browsing of the entire railway line.