For autonomous driving systems, pedestrian crossing detection is a crucial task. However, because it is dangerous to collect enough crossing data, the diversity and the scale of the crossing behavior are limited, while the data gathered from the real world has guidance role for the type definition of crossing behavior in virtual environment. In this paper, we use the open source software CARLA to synthesize the pedestrian crossing process based on the data of real pedestrian crossing data in the real world. When we build the synthetic scene, we enumerate the driving scene and also take random setting to enlarge the diversity of the pedestrian crossing types, and obtain 720 different scenes with 215,196 frames (named as Virtual PedCross-720). Then, we save the synthetic results in the format of COCO dataset for easy utilization. Finally, we use the YOLOv5 algorithm to train the Virtual PedCross-720 dataset, and we observe that the Virtual PedCross-720 dataset can form a stable pedestrian detection for the virtual and real data.
Virtual PedCross-720: A Synthetic Benchmark for Pedestrian Crossing Detection in Autonomous Driving Scenarios
Lect. Notes Electrical Eng.
International Conference on Autonomous Unmanned Systems ; 2021 ; Changsha, China September 24, 2021 - September 26, 2021
Proceedings of 2021 International Conference on Autonomous Unmanned Systems (ICAUS 2021) ; Kapitel : 90 ; 910-921
2022-03-18
12 pages
Aufsatz/Kapitel (Buch)
Elektronische Ressource
Englisch
AUTONOMOUS COMPLEX OF ARRANGEMENT OF PEDESTRIAN CROSSING
Europäisches Patentamt | 2021
|