Motivation
- As the emergence of autonomous driving technology�→ growing focus on pedestrian safety
- One of the key technologies to achieving these goals�: predict whether pedestrians will cross or not cross
- However, it is not easy! → The intention of humans is unclear!
Introduction
- To overcome these limitations, in this paper, we propose a novel fusion approach for anticipating pedestrian crossing intention
Summary
- We propose a novel feature fusion model, CIPF, which utilizes eight input modalities with three modules.
- We achieved performance outperforming the state-of-the-art methods on the PIE dataset with 91% prediction accuracy.
- We provide ablation studies and introduce the qualitative analysis of pedestrian crossing intention.
Ablation Studies
Qualitative Results :crossing :not-crossing
CIPF: Crossing Intention Prediction Network based on Feature Fusion Modules for Improving Pedestrian Safety
Je-Seok Ham1 Dae Hoe Kim1 NamKyo Jung2 Jinyoung Moon1
1Electronics and Telecommunications Research Institute (ETRI) 2Korea University
<Effectiveness of Input Features>
<Validity of Visual Encoder>
<Prediction Time Studies>