Absorption-Based, Passive Range Imaging from Hyperspectral Thermal Measurements

TL;DR

The paper tackles passive range imaging in natural LWIR scenes where object and air temperatures are similar. It develops a physics-based forward model that includes atmospheric absorption and air emission, and compares a bispectral baseline with a full hyperspectral inversion that jointly estimates distance, emissivity, and temperature using a smooth emissivity prior. A Fisher information analysis clarifies when and where distance information is available across the spectrum and identifies an optimal attenuation level to maximize information content, supporting the use of wide spectral data. Experimental results on 256-band LWIR data (8–13 µm) show range recovery from 15 to 150 m with emissivity and temperature maps, and a downwelling-detection mechanism based on the ozone line to flag unreliable pixels, achieving qualitative agreement with lidar for reliable regions.

Abstract

Passive hyperspectral longwave infrared measurements are remarkably informative about the surroundings. Remote object material and temperature determine the spectrum of thermal radiance, and range, air temperature, and gas concentrations determine how this spectrum is modified by propagation to the sensor. We introduce a passive range imaging method based on computationally separating these phenomena. Previous methods assume hot and highly emitting objects; ranging is more challenging when objects' temperatures do not deviate greatly from air temperature. Our method jointly estimates range and intrinsic object properties, with explicit consideration of air emission, though reflected light is assumed negligible. Inversion being underdetermined is mitigated by using a parametric model of atmospheric absorption and regularizing for smooth emissivity estimates. To assess where our estimate is likely accurate, we introduce a technique to detect which scene pixels are significantly influenced by reflected downwelling. Monte Carlo simulations demonstrate the importance of regularization, temperature differentials, and availability of many spectral bands. We apply our method to longwave infrared (8--13 $μ$m) hyperspectral image data acquired from natural scenes with no active illumination. Range features from 15m to 150m are recovered, with good qualitative match to lidar data for pixels classified as having negligible reflected downwelling.

Figures (15)

• Figure 1: Simulation of hyperspectral measurements in units of microflicks (µW ⋅ sr^-1 ⋅ cm^-2 ⋅ µm^-1) of a rock at 300 K from 10 m, 20 m, and 30 m ranges through the atmosphere at 289.7 K. Nearly all the absorption in this range is due to water vapor, which here has a volume mixing ratio (VMR) of 0.012. The instrumental spectral response is a Gaussian with 40 nm full width at half maximum. Zoomed area of the spectrum is highlighted with a dashed box. Atmospheric absorption depends on the wavelength and the range of the object.
• Figure 2: An example of the hyperspectral absorption-based ranging method of Section \ref{['sec:hyperspectral']} on experimental data: \ref{['fig:Teaser_Depth_Map']} depth map; \ref{['fig:Teaser_Emissivity_Clusters']} emissivity clusters; and \ref{['fig:Teaser_Temperature_Map']} temperature map. The range is extracted mostly from the absorptive spectral channels, with black pixels representing unreliable distance estimates due to significant downwelling contributions. Processing a wide spectral range enables object material identification. To show the value of the emissivity profile, $k$-means clustering is performed on the estimated emissivity profiles. The temperature is also estimated jointly, giving more information about the surroundings. The horizontal stripe artifacts in the results are due to data collection strategy with a scanning-based pushbroom sensor, as discussed in Section \ref{['sec:experimental_data_spec']}. \ref{['fig:Teaser_RGB_Image']} RGB image for visual reference of the scene.
• Figure 3: Conceptual figure for radiative transfer model. Blue and red arrows represent the contributions from object emission and air emission to the observed spectrum, respectively. The gray arrow represent the reflected thermal radiation from other objects such as the sky and the rock. Nearly all the absorption features in the LWIR band (8--13 µm) are due to water vapor. The attenuation coefficient of a model atmosphere with 0.012 volume mixing ratio of water vapor along with 289.7 K temperature and 1010 m bar pressure is plotted above.
• Figure 4: \ref{['fig:Fisher_information_Temperature_diff']} Simulation of the forward model for a rock at $-10$ K, $-5$ K, and +10 K relative temperatures with respect to air. The air temperature is 289.7 K, and the corresponding black-body radiance is plotted with a dashed red line. The atmosphere is at 1010 m bar pressure, and has 0.012 VMR water vapor content. The emissions from the object at different temperatures are plotted with dashed colored lines. Observed spectra at 30 m range are shown in solid lines of matching colors. Objects with higher temperature differences show bigger atmospheric features. \ref{['fig:Fisher_information_Spectrum_1']} Normalized Fisher information for a rock at 30 m, 70 m, and 200 m. The information profile is dependent on the true distance and spread over the LWIR spectrum around the water vapor absorption lines. \ref{['fig:Fisher_infromation_Spectrum_2']} shows the area in black box. For short distances, the most absorptive bands, in the 7--8 µm region, contain the most information. As distance increases, information shifts to longer wavelengths. Most of the information in the collected measurement band (8--13 µm) is in 8--9 µm.
• Figure 5: Bispectral ranging using one clear (8.38 µm) and one absorptive band (8.42 µm). Depth maps for \ref{['fig:Air_emission_Neglecting_air']} neglecting and \ref{['fig:Air_emission_Including_air']} including the air emission. \ref{['fig:Air_emission_lidar']} Depth map produced by lidar for comparison, without pixel-to-pixel correspondence with the hyperspectral sensor; black pixels represent missing data. Note that the color bar range is different for the first plot because most of the range estimates are negative or close to zero when air emission is neglected. \ref{['fig:Air_emission_Temperature_difference']} Estimated temperature difference with respect to air, calculated using brightness temperature at the clear band for object temperature and weather station data collected on-site for air temperature.