The Cerema pedestrian database: A specific database in adverse weather conditions to evaluate computer vision pedestrian detectors

Abstract

Nowadays, many pedestrians are victims of road accidents. Several artificial vision solutions, based on pedestrian detection, have therefore been developed to assist drivers and reduce the accident rate. But most of the proposed pedestrian databases make it possible to test detection only in favorable conditions. The main goal of this research is to provide a learning and testing environment for the development of pedestrian detectors able to function under all weather conditions by day and even by night. This paper presents a new database, called Cerema, composed of 10 sets which include normal and degraded conditions (day, night, fog, rain). Image data will include detailed annotations for each set. Two common detectors are used to show the usefulness of our database, which are HOG and Haar. Finally, the results obtained on this new database will be presented to show the impact of adverse weather conditions on these two different detectors.