An Omnidirectional Stereo System for Logistic Plants. Part 1: Calibration and Multi-channel Rectification

This work presents a fisheye lens based omnidirectional stereo sensor. The sensor is installed on the top of the vehicle such that it can observe the entire area around it. The vehicles we use in the experiments are automated forklifts. We propose a multi-channel rectification to divide the original fisheye images into three pairs of rectified perspective images. The complexity of the fisheye stereo processing therefore reduces to a series of standard image processing techniques. This is especially useful for the efficient generation of the 3D information from images, where the epipolar lines have to correspond to the lines in the images. We use a standard dense stereo matching engine to generate three disparity images from the pairs of rectified images. The stereo reconstruction, the processing of the disparity images for obstacles detection and the detection accuracy evaluations are presented in the second part of the paper.