Researchers have developed a novel two-stage system for detecting and tracking hidden dynamic objects using audio representations. The first stage involves self-supervised pre-training on raw audio waveforms, inspired by the Joint-Embedding Predictive Architecture (JEPA), to predict future audio segments from past context. This is followed by supervised multi-task fine-tuning using a bidirectional LSTM with three classification heads to estimate the number of vehicles, their types, and their direction of arrival. The system was evaluated on a newly collected dataset and demonstrated superior performance compared to existing methods, showing robust representations transferable to unseen driving scenarios. AI
IMPACT This research could enhance the safety of autonomous vehicles by improving their ability to detect and track hidden objects using audio cues.
RANK_REASON The cluster contains a research paper detailing a new AI methodology. [lever_c_demoted from research: ic=1 ai=1.0]
- alphaXiv
- arXiv
- CatalyzeX
- DagsHub
- Gotit.pub
- Hugging Face
- Katerina Vinciguerra
- long short-term memory
- ScienceCast
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
