Understanding traffic scene images taken from vehicle mounted cameras provides important information for high level tasks such as autonomous driving and advanced driver assistance. The problem is hard due to challenges from weather and illumination variation. To facilitate the research against such challenges, in this paper we present a new benchmark for cross-weather traffic scene understanding1. The dataset consists of 1,356 traffic scene images collected at 226 different locations. For each location, there are six images taken by a vehicle mounted camera under different weather/illumination conditions including sunny day, night, snowy day, rainy night and cloudy days. We manually annotated each image with scene understanding labels such as road, sky, building, etc. To the best of our knowledge, this is the first carefully collected benchmark for cross-weather traffic scenes. In addition, we also provide results from two popular scene parsing systems as the baselines. We expect the benchmark to help boost research in improving robustness of traffic scene understanding algorithms.
A benchmark for cross-weather traffic scene understanding
01.11.2016
1687966 byte
Aufsatz (Konferenz)
Elektronische Ressource
Englisch