Edge computing for driving safety: evaluating deep learning models for cost-effective sound event detection

This paper addresses road safety concerns by investigating low-cost solutions for sound event detection (SED) tailored to driving scenarios. While advanced tech- nologies like deep learning hold promise for improving road safety, their practi- cal implementation often involves expensive sensors and hardware. Distractions, a major cause of accidents, require effective detection and mitigation. This study con- centrates on auditory distractions and utilizes SED with low-cost edge devices to identify and timestamp relevant audio events, providing valuable insights into the driving environment. We evaluate state-of-the-art deep learning models on various edge devices, including the 2023 DCASE baseline with convolutional recurrent neu- ral networks (CRNN) and an adapted YOLO vision model for audio spectrograms. Our analysis spans different hardware options, including single-board computers (SBCs) and desktop equipment, offering guidance on cost-effective hardware selec- tion for in-vehicle SED applications. This research aims to contribute to affordable SED solutions in the context of driving safety, with the ultimate goal of advancing road safety efforts worldwide.