Novel Index for Objective Evaluation of Road Detection Algorithms

Road detection is a relevant task within vision-based systems devoted to assist the driver. Although they have been improved during the last decade, these algorithms are usually validated using qualitative results. Nonetheless, quantitative evaluation is necessary either to enable the comparison between different algorithms or to achieve the optimal performance of a given one. In this paper we present a composite index to quantitatively assess the performance of road detection algorithms. The measure is based on a weighted combination of different evaluations which use a trade-off between precision and recall scores. Obtaining a single index score is a major benefit. It can be used to easily compare algorithms or to properly set their parameters. Moreover, innovatively our proposal includes a human perception criterion to improve its usefulness. Experiments on real-world data corroborate the usefulness of the proposed index.