Navy

Autonomous water flow characterization from surface water velocity

Using cameras for remote field-scale flow measurement and navigation

Sensors Environmental

team of U.S. Navy scientists working at the Naval Research Laboratory have recently developed a technique to remotely characterize water flow using one or more camera devices that include sensors, a feature tracker, and a 3D displacement generator. Depending on the scenario, cameras can be mounted to stationary objects (e.g., a bridge, sensor station) or vehicles (e.g., unmanned aerial vehicle, water vessel). The cameras are configured differently depending on their number. The patented technology is available via license agreement to companies that would make, use, or sell it commercially. 

Depth estimations in uncharted rivers would enable safe navigation and flood hazard predictions. Measurements of field-scale flows are required for accurately understanding, modeling, and predicting the dynamics in riverine and marine environments. Existing techniques to accurately capture field-scale flow measurements for both typical and hazardous flow conditions include turbine flowmeters and ultrasonic profilers 

While ultrasonic meters provide precise flow measurements, they require substantial installation infrastructure and cannot be rapidly placed for time-critical deployments. In addition, the meters are not robust enough to be used in hazardous flow conditions (e.g., violent mudslides, flash floods, or debris flows). For such extreme conditions, remote sensing flow measurement techniques are required. Particle Image Velocimetry (PIV) is a widely-used, non-contact, image processing laboratory technique that utilizes cross-correlation of consecutive images via Finite Fourier Transform (FFT). Large Scale Particle Image Velocimetry (LSPIV) extends the laboratory technique using FFT to correlate image pairs in field measurements. However, LSPIV is limited due to its time consuming and demanding user input requirements.  

In response, NRL scientists have discovered a way to obtain the velocity field of a stream surface without user input. Images captured by a specialized camera can be cross-correlated (e.g., features from accelerated segment test (FAST) algorithm) to automatically extract image subregions rich in color texture and detect trackable regions (i.e., features). Feature matching is achieved by comparing features (e.g., binary robust independent elementary features (BRIEF)) along with enhanced subpixel displacement detection based on, for example, the Lucas-Kanade optical flow equation. To increase efficiency, the nearest neighbor search is performed to restrict the features compared to only those that are close to each other.  

Images are obtained at regular intervals so that time-series analyses can be performed. An image pyramid is generated for each image and can include various levels of images of progressively reduced length-scale. For example, the first level image can be at the original scale, the second level image can be at half the original scale, the third level image can be a quarter the original scale, and so on. Various scale reductions and numbers of levels can be used for the image pyramid. 

Features are detected, described, matched, and tracked at the pixel-level across the time-series of images. The tracked features result in displacements (i.e., vectors) that show the magnitude and direction of the features during each time step.  

By tiling and normalizing the grey-scale of each image with a heuristically defined length-scale, features can be extracted while significantly suppressing the effect of shadows and steep grey-scale/color gradients, which normally would lead to the detection of a very low number of features. The size of the tile can be preconfigured or determined automatically based on the size of the original image. Tile size remains consistent across the different levels of the image pyramid, and the number of levels in the pyramid can be dependent on the tile size. 

A navigation component is used to control the watercraft. For example, water flow characterization can be used to assist in avoiding collisions while navigating the watercraft at river location. In another example, the navigation component can track the current location of the watercraft so that the location, as determined by a global positioning system (GPS) can be rendered in a 3D model of the water flow characterization. 

Do you have questions or need more information on a specific technology? Let's talk.

Contact Us