ETRI-Knowledge Sharing Plaform

KOREAN
논문 검색
Type SCI
Year ~ Keyword

Detail

Conference Paper CIPF: Crossing Intention Prediction Network based on Feature Fusion Modules for Improving Pedestrian Safety
Cited 8 time in scopus Download 87 time Share share facebook twitter linkedin kakaostory
Authors
Je-Seok Ham, Dae Hoe Kim, NamKyo Jung, Jinyoung Moon
Issue Date
2023-06
Citation
Conference on Computer Vision and Pattern Recognition Workshops (CVPRW) 2023, pp.3665-3674
Language
English
Type
Conference Paper
DOI
https://dx.doi.org/10.1109/CVPRW59228.2023.00374
Project Code
23HS6500, Development of Previsional Intelligence based on Long-term Visual Memory Network, Moon Jin Young
Abstract
As the development of autonomous driving technology continues, pedestrian safety is becoming an increasingly important issue. The ability of an autonomous car to accurately predict whether a pedestrian will cross the road is essential for ensuring their safety, as the vehicle can slow down in time or stop to avoid any potential accidents. However, predicting pedestrian behavior is a complex task influenced by various environmental and contextual factors. To deal with this issue, we propose a novel method, Crossing Intention Prediction based on feature Fusion modules (CIPF) that combines eight different input features extracted from both pedestrians and vehicles through three fusion modules using RNN layers and attention mechanisms. We demonstrated state-of-the-art performance of prediction accuracy in the PIE dataset, which is the most widely used for pedestrian crossing intention prediction. We also demonstrated the superiority of the performance of our CIPF network through qualitative and quantitative analysis. In particular, we also performed ablation studies on the verification of the effectiveness of the eight input features, the validity of VGG encoders, and performance comparison of our CIPF over time by adjusting the prediction time.
KSP Keywords
Art performance, Attention mechanism, Autonomous car, Contextual factors, Feature Fusion, Input features, Intention prediction, Over time, Pedestrian safety, Performance comparison, Performance of prediction