Improving pedestrian detection using MPEG-7 descriptors

Abstract. During the last decade, modern Pedestrian Detection Systems made massive use of the steadily growing numbers of high-performance image acquisition sensors. Within our naturalistic driving environment, a lot of different and heterogeneous scenes occur that are caused by varying illumination and weather conditions. Unfortunately, current systems do not work properly under these hardened conditions. The aim of this article is to investigate and evaluate observed video scenes from an open source dataset by using various image features in order to create a basis for robust and more accurate object detection.