Accuracy of free-space detection: Monocular versus binocular vision

Autonomous on-road vehicles or vision-based driver assistance benefit from free-space analysis. This paper evaluates the accuracy of free-space detection in stereo and monocular vision on KITTI benchmark data. Such an evaluation of low-level computer vision algorithms is, for example, also necessary as free-space analysis is recently becoming an important module for designing vehicle test beds. The novelty of this paper is defined by comparing a designed monocular algorithm with a selected binocular algorithm on long sequences of images, taken for different road profiles and lighting conditions. The results demonstrate potentials of monocular vision as applicable, for example, if using mobile devices only. Furthermore, this paper extends the detection of a lower envelop in a v-disparity image, usually done by estimating a straight line or B-spline, by using polynomial curve fitting.