Absorption-Based, Passive Range Imaging From Hyperspectral Thermal Measurements.

Prior research has shown that passive hyperspectral longwave infrared measurements provide extensive data regarding environmental surroundings through the detection of thermal radiance. The specific material composition and surface temperature of a remote object determine the initial spectrum of thermal energy emitted into the atmosphere. As this energy travels toward a sensor, the spectrum undergoes significant modification based on the distance traveled, ambient air temperature, and local gas concentrations. Previous methodologies for distance estimation often required the presence of hot, highly emitting objects to distinguish the target from the background thermal noise. Ranging becomes exceptionally challenging when the temperature of the observed objects does not deviate significantly from the surrounding air temperature. This difficulty arises because the thermal contrast between the object and the atmosphere is minimized, complicating the spectral separation process. This absence of evidence motivated the development of a computational framework capable of separating intrinsic object properties from atmospheric propagation effects.

This research introduces a novel passive range imaging method that computationally separates thermal radiance from atmospheric propagation effects to determine distance. The approach jointly estimates the range and intrinsic properties of an object while providing explicit consideration for the emission of the intervening air. To address the underdetermined nature of the inversion process, the researchers employed a parametric model of atmospheric absorption. The study incorporates regularization techniques designed to produce smooth emissivity estimates, which helps stabilize the mathematical recovery of object features. A specialized detection technique was also developed to identify scene pixels that are significantly influenced by reflected downwelling radiation from the sky. This investigation specifically evaluates how temperature differentials and the availability of numerous spectral bands influence the accuracy of the resulting range estimates. The primary goal is to enable accurate ranging even when the temperature of the target is nearly identical to the ambient environment.

The team utilized Monte Carlo simulations to analyze the necessity of regularization and the impact of temperature differentials on the estimation process. They applied the computational framework to Longwave Infrared (LWIR) hyperspectral image data spanning the 8-13 μm spectral range. Natural scenes were captured using passive sensors without any active illumination to validate the method under realistic environmental conditions. The algorithm classifies each pixel to determine if reflected downwelling is negligible, which is a prerequisite for accurate range recovery in this model. Researchers compared the recovered range features, which spanned from 15 m to 150 m, against ground-truth measurements obtained from Light Detection and Ranging (lidar) systems. The mathematical inversion process relies on a parametric model to handle the complexities of atmospheric absorption and gas-specific spectral signatures. This methodological approach ensures that the intrinsic emissivity of the object is recovered alongside the distance measurement to provide a complete scene profile.

The proposed method successfully recovered range features from distances between 15 m and 150 m using only passive thermal measurements. Pixels identified as having minimal reflected downwelling showed a high qualitative match to the data provided by the lidar reference system. Monte Carlo simulations confirmed that regularization is essential for maintaining accuracy, especially when the temperature difference between the object and the air is small. The availability of numerous spectral bands significantly improved the reliability of the joint estimation process by providing more data points for the inversion. The technique effectively separated the intrinsic emissivity of the object from the atmospheric modifications caused by propagation through the air. The study demonstrated that the 8-13 μm spectral window contains sufficient information for passive distance estimation in complex natural environments. These results indicate that the computational model can overcome the limitations of low thermal contrast in outdoor scenes without requiring active light.

These findings suggest that absorption-based passive imaging can provide reliable depth information without the need for active light sources or lasers. The ability to estimate range for objects near ambient temperature expands the utility of thermal sensors in diverse environmental and surveillance applications. Future research may focus on refining the parametric models to account for more complex atmospheric gas concentrations and varying humidity levels. This methodology offers a potential alternative to active ranging systems in scenarios where covert operation or low power consumption is required. The integration of smooth emissivity regularization provides a robust framework for improving other hyperspectral inversion tasks in remote sensing. The researchers conclude that identifying reflected downwelling is a critical step for ensuring the accuracy of passive thermal ranging in real-world scenes. This study establishes a foundation for future developments in passive depth sensing using longwave infrared technology for autonomous navigation.