BlackIceNet: Explainable AI-Enhanced Multimodal for Black Ice Detection to Prevent Accident in Intelligent Vehicles
Paper Title: BlackIceNet: Explainable AI-Enhanced Multimodal for Black Ice Detection to Prevent Accident in Intelligent Vehicles
Journal Name: IEEE Internet of Things
Abstract: The advancement of intelligent transport systems and the emergence of autonomous cars have the potential to reduce road accidents significantly. However, mountainous regions, like South Korea, are especially vulnerable to the threats of black ice formation in winter. Due to its near-invisibility and sudden formation, this translucent layer of ice is challenging to detect. It poses a severe risk to both human and autonomous drivers. Conventional approaches to monitoring road conditions frequently struggle to swiftly identify black ice, highlighting the need to advance more sophisticated sensing systems. This project aims to identify road surface conditions, including black ice, by integrating images and audio with sensor data, such as surface and ambient temperatures. The system utilizes convolutional neural networks (CNNs)-based multimodal system to analyze visual and acoustic data to detect road surface conditions. The data was gathered over two years from a testbed set up in three separate terrains in South Korea. Subsequently, drivers are promptly notified of potential dangers on the road, thereby reducing the probability of accidents resulting from hazardous conditions. The evaluation results indicate the efficacy of the suggested method, as the multimodal system achieves a 97\% accuracy rate in detecting black ice. The multimodal fusion system overcomes individual techniques' limitations, providing dependable early warnings for black ice and improving road safety.