Linear Protection Grid Optimized Image Stitching for Mobile Robots

Image stitching can be used to in 3D reconstruction to obtain the comprehensive obstacle information, which plays an important role in the field of mobile robots. However, previous algorithms have two problems: 1. The linear structure of the image might have been corrupted. 2. Some inconsistency may exist in the transitional region of the stitched image. In order to solve above problems, in this paper, we propose a grid-based linear structure protection method, which applies the constraints to the lines extracted from the image to protect them from the distortion caused by the mesh deformation process, and resulting in a natural panorama with reduced distortion. This method helps to obtain a natural panoramic image with reduced distortion. At the same time, we use the neighbor weighted based boundary artifact removal approach to process the stitched image, which can avoid the stitching problem and can make the image look more natural. We conducted some experiments, and the results demonstrated that our method is more efficient and more natural as compared with some state-of-the-art methods.

I. Introduction

The main way robots get information is the same as humans, mainly through vision. When a mobile robot completes complex, cumbersome, and dangerous tasks, the robot vision system must maintain very high stability and realtime performance. There is no doubt that robot vision must be a difficult and hot spot in mobile robot research. What is different from the human eye is that humans have only one pair of eyes. The information about the external environment is limited by the body structure. The robot can have multiple pairs of eyes at the same time as its processor processing power allows, so that so many eyes can provide robots with a full range of information from the main body and generating panoramic images has become a key technology.