Non-line-of-sight vehicle detection via audio-visual fusion

Wang, Huaxuan, Yu, Huilong, Zhang, Ruizeng, Zhou, Wei and Xi, Junqiang 2026. Non-line-of-sight vehicle detection via audio-visual fusion. Presented at: ICASSP 2026 - 2026 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), Barcelona, Spain, 3-8 May 2026. ICASSP 2026 - 2026 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP). IEEE International Conference on Acoustics Speech and Signal Processing. IEEE; 1999, pp. 11807-11811. 10.1109/icassp55912.2026.11465095

Full text not available from this repository.

Abstract

Acoustic signals exhibit great potential for Non-Line-of-Sight (NLoS) perception owing to their ability to traverse obstacles, while scene structures play a critical role in shaping sound reflection and diffraction. However, existing studies rarely account for these features, making the accurate perception of NLoS targets challenging. In this work, we propose a scene-aware acoustic perception network that integrates audio and visual signals for occluded vehicle detection. The network leverages bird's-eye view (BEV) images to represent scene spatial features and derives acoustic information from time-frequency and spatio-temporal spectra. A combination of CNN, LSTM, and Conformer modules is employed to process multimodal signals, with the Conformer enhancing global-local feature modeling and substantially improving occluded vehicle detection performance. Experimental results demonstrate that our algorithm outperforms the state-of-the-art methods and achieves accuracies of 94.1% and 97.0% in OVAD and AOVD datasets, respectively.

Item Type:	Conference or Workshop Item - published (Paper)
Date Type:	Publication
Status:	Published
Schools:	Schools > Computer Science & Informatics
Publisher:	IEEE; 1999
ISBN:	979-8-3315-6702-6
ISSN:	1520-6149
Last Modified:	08 May 2026 09:15
URI:	https://orca.cardiff.ac.uk/id/eprint/186871

Actions (repository staff only)

Edit Item