Homography-Based Industrial Image Stitching

Global robotic vision on large-scale workspaces frequently requires the composition of data from different or moving sensors. We propose a method for constructing accurate panoramic images of a robotic workspace from standard 2D images by estimating the trajectory of a moving camera. Utilizing the camera’s trajectory directly yields the positional homographies into the panoramic image, without the need to extract features and works even in the presence of zero overlap. Finally, we evaluate the overall performance of the panoramic vision approach in combination with a robotic setup on a flat work plane.