A Framework for Background Modeling Using Vehicle-to-Infrastructure Communication for Improved Candidate Generation in Pedestrian Detection
Ghaith Al-Refai, Modar Horani, Osamah A. Rawashdeh · 2018
Vision-based pedestrian detection is an essential part of Advanced Driver Assistance Systems (ADAS). A pedestrian detection system involves several processing steps including image acquisition, candidate generation, classification, and realtime tracking. Typical approaches for pedestrian candidate generation scan the whole image, which is time consuming. Intelligent generation of potential pedestrian candidates, by reducing the number of the unnecessary candidates, may improve the detection accuracy and reduce the runtime of detection algorithms. This paper introduces a new framework for improved candidate generation using background modeling. The proposed method leverages Vehicle-to-Infrastructure (V2I) communication for sharing of image frames. The system architecture is introduced and defined in-terms of requirements and tasks. Promising initial results for background modeling and moving object detection are provided with discussion and suggestions for future work.