System and method for visual anemometry using flow-structure interactions

Abstract

Systems and methods for determining wind speed are disclosed. The system includes a video capture device (e.g., a smartphone camera, a drone-mounted camera, or a satellite) to capture a video time-series of a pre-existing flexible object, such as natural vegetation, interacting with a wind flow. A processor is configured to analyze the video time-series to extract kinematic features of the object's motion and determine a quantitative wind speed. In one embodiment, a physics-informed neural network is used, which is trained to infer the object's mechanical properties (such as stiffness or species) from the video to solve for the wind speed. In another embodiment, a neural network is configured to receive a set of statistical feature maps, derived from the video's temporal signals, to determine the wind speed. In a further embodiment, the wind speed is determined by applying a first-principles physical formula based on the discovery that the fluctuating motion of individual leaves becomes decoupled from the branch structure at certain wind speeds, allowing for a quantitative measurement based on leaf size and measured leaf speed.

Dependencies are inferred from the claim text.