POWER OPTIMIZED CO-REGISTRATION OF MULTISPECTRAL AND THERMAL CAMERAS FOR REAL-TIME URBAN PLUVIAL FLOOD MONITORING

Urban pluvial flooding occurs when precipitation intensity exceeds the capacity of drainage systems, causing significant risks to urban infrastructure and safety. While stream gage networks and flood risk maps are vital for managing coastal and riverine flooding, there remains a critical gap in real-time monitoring and flood risk assessment for urban pluvial flooding. Early detection and monitoring are essential for timely response and mitigation.

To address this challenge, our team is developing a low-power distributed sensor network with a camera-based flood mapping platform known as the Urban Flood Observation Network (UFO-Net). This system provides real-time monitoring and automatic high-water mark mapping. It contains two cameras: a multispectral camera (capturing red, green, blue, and near-infrared radiation) and a long-wave infrared (LWIR) thermal camera. The multispectral camera differentiates land cover types and water bodies, while the thermal camera identifies temperature changes associated with water. This research proposes a critical component: the development of a rapid, replicable workflow for co-registering optical and thermal camera fields of view in a power-constrained environment. This co-registration ensures precise data fusion and enhances the system's performance, reliability, and utility for flood monitoring and management.

To achieve these goals, we employ SimpleITK, a tool for iterative optimization of image transformation. Our final goal is to generate five-band images (RGB, NIR, LWIR) directly on edge using a Raspberry Pi 4 to classify inundated versus non-inundated areas, developing a bitmap for flood mapping. Additionally, we want the algorithm to adapt to changing sensor conditions and movement and be computationally and power-efficient. This optimized registration workflow supports developing a robust, real-time flood detection system tailored to urban environments.