Low-Power BLACK-ICE Detection for Safety Critical Edge Devices on Roads

The technological advancements in video surveil-lance systems represent a modern, secure, and sustainable in-frastructure for smart cities. Nevertheless, detecting anomalous events in surveillance footage remains challenging, often necessi-tating considerable human intervention. A notable anomaly with severe implications is the presence of BLACK-ICE on highway surfaces. The presence of BLACK-ICE renders the road slippery, heightening the risk of safety incidents for pedestrians and vehi-cles. Notably, identifying BLACK-ICE is challenging due to its transparent nature, distinguishing it from other slippery surfaces like wet or snowy roads. In this regard, Closed-Circuit Television (CCTV) systems deployed on roadways emerge as a valuable tool for BLACK-ICE detection. This research specifically focuses on investigating BLACK-ICE surface detection using CCTVs. A novel method based on Convolutional Neural Networks (CNN), which calculates the sharpness and glossiness factors of the road surface, has been introduced to enhance performance and optimize resource utilization without compromising accuracy.