Single-View Polarimetric Long-Range Maritime Target Shape Recovery

Etegent will achieve Navy 3D polarization requirements using polarimetric imagery and existing SfP algorithms. Etegent will address ocean surface-based collection scenarios where targets of interest are non-cooperative. Etegent will focus on SfP for detecting ships a few meters on a side, beyond one km, and at the horizon. Phase I work will demonstrate the ability to estimate target shape from polarimetric imagery using SfP techniques. The system will enable improved DRI of small maritime targets at ranges beyond one kilometer: distinguish non-background objects, separate maritime targets of interest and non-targets, and categorize maritime targets. Improved detection helps answer which objects in the image are manmade such as buoys, floating trash, and maritime vessels and which are natural environment such as sea, sky, and clouds. Improved recognition helps answer which non-background objects are maritime targets of interest, such as boats. Improved identification helps answer which targets of interest are adversarial and which are not threats. The Etegent team will collect data of RC model boats and real boats on lakes. Candidate collection sites are a few hours drive from the laboratory facility amd advantageous for controlling times of day and weather conditions. Lake collections can be conducted from land looking out to sea. Lake collections allow increasing range beyond one kilometer and view targets located at the water-horizon boundary. For example, lake freighters have lengths up to 1000 feet and beams of 100 feet, corresponding to a range of 43.5 km scaled to the nominal target at one kilometer. Lake ferries provide another collection opportunity. The ferries are 90 feet long with beams of 40 feet, and at 3.9 km equivalently scales to a nominal target at one kilometer. The lake ferries and freighters allow Etegent to evaluate SfP performance as a function of air path transmission and Rayleigh scattering and depolarization while maintaining a meaningful number of pixels on target.  The DoLP and Fresnel equations are used to estimate zenith angle. However, zenith ambiguity exists, with a value on each side of the Brewster angle, when using specular reflection relationship. Two polarimetric spectral bands disambiguate the specular zenith ambiguity, relying on radiometrically calibrated and well registered band images. Alternatively for monochromatic cameras, the diffuse relationship can be used, which is monotonic and lacks zenith ambiguity. The refractive index of the reflecting surface must be known to solve for zenith angle. Etegent will use normal reflection measurements and engineering tables to determine refractive index for the articles under test. The azimuth angle is inferred from AoP, however there is azimuthal ambiguity; the number of possible values equals the number of polarization filters used to generate AoP. The Etegent team will use published strategies to determine specular and diffuse contributions and resolve ambiguity.