A better software for estimating wind fields
A computer-vision program born at Inria, Typhoon is meant to retrieve vector motion fields from image sequences of turbulent flows such as weather satellite photos. After being used by California State University, Chico in conjunction with lidar imagery to assess low-altitude wind fields, this innovative software is now reaching the industry. At stake: finding the best spots for future wind farms or estimating the atmospheric dispersion of pollutants.
“In a nutshell, Typhoon is designed to observe and measure fluid flows from image sequences, says Pierre Dérian , the scientist who authored the software during his Ph.D. thesis in the Fluminance research team at Inria, in Rennes, Brittany, France. It belongs to a class of tools developed over the last 30 years for people working on image-based weather forecast, oceanography, aerodynamics, so on and so forth. Although a fluid flow per se cannot be seen, there is a variety of visible markers such as dust, smoke or clouds that are being transported by this flow. Their observable displacements provide indication as to the motion of the flow. Our software automates the recovery of these fluid velocity fields through visual analysis. ”
The novel algorithm comes in the wake of previous tools designed within Fluminance by scientists Etienne Mémin and Thomas Corpetti
. “The approach they initiated is a departure from the usual techniques used by most commercial applications including the Particle Image Velocimetry systems (PIV) that are commonly found in laboratories. The classical and so-called cross-correlation method consists in observing one tiny patch in an image and matching this patch again in the next image of the time series. This results in a first vector corresponding to a fluid displacement. Covering a whole field calls for many vectors, hence many patches,
” which is not a problem in and of itself as “the task is easily parallelizable.
” However, “if there happens to be a lack of information locally in the image, this method can't cope with it. In addition, it suffers from a certain imprecision, returning estimated velocity vectors on a grid coarser than input image data.
Lastly, if it works fine for in-lab particle measurement, it does not fare so well with outdoor scalar images, i.e. colourants, smoke, brine, clouds, etc.
By contrast, “instead of dealing with a string of local patches, the so-called ‘dense’ approach encompasses the whole field —and all the vectors— at once. It provides a displacement vector at every grid point.” The advantage? “It can overcome the lack of information at places based upon what's happening in the vicinity. ”
Typhoon adopts this technique but pushes the envelope further by introducing a mathematical tool known as wavelets. What is this about? “Well, all motion estimation methods feature two main aspects. One is the data model meant to link the estimated motion to image data. The other one is a regularization term designed to compensate for the local lack of information from images. Say a cloudless sky or a uniformly grey sky for instance. Such low textured areas are hard to process. Classically, spacial smoothing terms are employed. In Typhoon, the vector motion field is represented with wavelets, and that's the software's main particularity. Here, the wavelet formalism simplifies the design of the regularization terms. Another advantage is that wavelets form intrisically multiscale bases, which echoes to the multiscale nature of fluid dynamics. The flow recovery is then comparable to a Google Map, where details are progressively added as one zooms in. ”
Measuring the Displacement of Pollution
Over the last three years, the software has been deployed in the context of lidar imagery. Located at CSU, Chico, California and managed by scientist Shane Mayor
, the REAL is a lidar designed to capture the displacement of aerosol particles such as dust, smoke, pollen or pollution resulting from human activities. Housed in a shipping container on a 40-foot long trailer, this laser beam is a tool of choice for estimating wind displacements in the aerosol-rich lower atmosphere where satellite imagery proves of little avail as clouds usually cruise at higher altitudes. During a series of experiments, a study has shown that Typhoon performs better than the cross-correlation techniques as it resolves finer spatial-scale flow details
Although Typhoon is still at prototype stage, two commercial licenses have been recently granted. The first customer is Spectral Sensor Solutions (S3) LLC, a Virginia-based startup company working on REVEAL, a compact lidar mounted in the back of a pickup. The second one is Space Dynamics Laboratory (SDL), a unit of the Utah State University Research Foundation that displays similar interest in lidar-assisted wind field estimation.
Having said that, Typhoon may also be of service in completely different contexts. “The Geomorphology and Sediment Transport Laboratory of the United States Geological Survey (USGS) has also acquired a license. They are conducting research on floods, flying drones over the rivers and using Typhoon to extract surface flows from infrared imagery. They haven't published yet, Dérian remarks, but we are eager to see how the software performs. ” An experimental demo version is now available as a web app on AllGo, which is an Inria's platform meant to facilitate the use of the scientific applications.
These articles could interest you:
Typhoon is co-owned by Inria and the CSU, Chico Research Foundation. Licensing is handled by both institutions. Licenses are free for public research activities.