Heavy rain disrupts photogrammetry by introducing moving droplets, lens contamination, and reduced contrast that break correspondence and depth estimation. Causes include rapid specular streaks from falling raindrops, water films on optics that refract and blur light, and atmospheric scattering that lowers signal to noise. The consequences are lower point density, biased surface normals, and mapping gaps that matter for disaster response, infrastructure inspection, and ecological surveys across flood-prone regions. Research by Shree K. Nayar at Columbia University has characterized how rain alters image formation, informing both physical countermeasures and image-restoration algorithms. Field robotics work led by Sebastian Scherer at Carnegie Mellon University emphasizes operational safety and autonomy under adverse weather, guiding mission-level decisions for when to collect, pause, or adapt sensing strategies.
Sensor fusion and hardware strategies
Robust autonomous photogrammetry in heavy rain depends on combining redundant sensors and ruggedized optics. Polarizing filters and hydrophobic or heated lens covers reduce accumulated droplets on critical apertures, while structural radomes and wipers protect cameras during sustained exposure. Active sensors such as imaging LiDAR and short-range radar provide geometric returns less affected by optical streaks, and inertial measurement units combined with RTK GNSS preserve pose when visual tracking fails. Davide Scaramuzza at University of Zurich has shown the practical value of event-based vision and visual-inertial fusion for resilience to difficult visual conditions, an approach that mitigates motion blur and transient artifacts common in rain.
Software and learning-based corrections
Onboard processing must detect degraded frames and apply robust matching and restoration before triangulation. Classical approaches use temporal median and multi-view consistency to reject transient raindrop features, while modern methods leverage deep de-raining networks trained on real raining imagery to reconstruct underlying textures. Nuance matters because aggressive deraining can hallucinate details that bias measurements; therefore confidence scoring and uncertainty propagation into the 3D reconstruction pipeline are critical. Autonomous planners should increase image overlap, slow platform velocity, and alter altitude to reduce droplet angular velocity relative to the sensor, trading mission speed for data quality.
Operational deployment balances human needs and environmental context: in tropical coastal communities reliable mapping during monsoon events supports humanitarian logistics, while minimizing disturbance to sensitive habitats is essential. Combining proven physical protections, complementary sensors, adaptive autonomy, and validated image-restoration methods yields high-quality photogrammetry in heavy rain without relying on single-point solutions.