Surrounding View for Enhancing Safety on Vehicles

This paper presents a vision system based on several wide angle cameras and robust methods to compose a view of the vehicle perimeter in only one image. Our proposal can be split into two stages: firstly, an offline calibration step and secondly, an online panorama composition given the acquired images from the cameras. The images are horizontally undistorted, warped and stitched together according to lookup tables (LUTs) generated during the calibration process. The offline part of the method calibrates every camera individually and computes the pixel mapping required for the images transformation into the surrounding view. This mapping encodes a special undistortion process and the image registration between adjacent cameras. In particular, the proposed method does not fully undistort the images but it projects them onto a cylindrical surface to reduce the amount of radial distortion and facilitate the panoramic stitching of the images. Our approach includes a feathering algorithm for seams removal and a polar conversion of the panoramic view. This method has been successfully tested on a small wheeled robot and it is fully scalable to larger platforms such as an armored patrol. Vehicles safety can be enhanced thanks to perimeter monitoring.