A Hardware-Algorithm Co-Designed Framework for HDR Imaging and Dehazing in Extreme Rocket Launch Environments

TL;DR

This work addresses the challenge of extracting accurate mechanical parameters from rocket plume imagery under extreme haze and illumination by introducing a hardware–algorithm co-design that couples a Spatially Varying Exposure (SVE) sensor with a data-driven haze perception and adaptive multi-exposure fusion. The framework captures multi-exposure data in a single shot, estimates haze density without atmospheric priors, and performs region-aware, entropy-constrained fusion across scales to recover radiance faithfully. Key contributions include a specialized SVE camera with a 2×2 macro-pixel attenuator, a four-feature haze perception model (BI, WC, CF, V) with region-based enhancement, and an adaptive pyramid fusion that preserves structure while suppressing haze. Extensive experiments on on-site launches, laboratory combustion, and simulated hazy scenes demonstrate superior dehazing, HDR recovery, and quantitative utility for combustion diagnostics and particle-velocity measurements, highlighting significant practical impact for extreme aerospace environments.

Abstract

Quantitative optical measurement of critical mechanical parameters -- such as plume flow fields, shock wave structures, and nozzle oscillations -- during rocket launch faces severe challenges due to extreme imaging conditions. Intense combustion creates dense particulate haze and luminance variations exceeding 120 dB, degrading image data and undermining subsequent photogrammetric and velocimetric analyses. To address these issues, we propose a hardware-algorithm co-design framework that combines a custom Spatially Varying Exposure (SVE) sensor with a physics-aware dehazing algorithm. The SVE sensor acquires multi-exposure data in a single shot, enabling robust haze assessment without relying on idealized atmospheric models. Our approach dynamically estimates haze density, performs region-adaptive illumination optimization, and applies multi-scale entropy-constrained fusion to effectively separate haze from scene radiance. Validated on real launch imagery and controlled experiments, the framework demonstrates superior performance in recovering physically accurate visual information of the plume and engine region. This offers a reliable image basis for extracting key mechanical parameters, including particle velocity, flow instability frequency, and structural vibration, thereby supporting precise quantitative analysis in extreme aerospace environments.

Figures (10)

• Figure 1: Hardware-algorithm co-design framework diagram. (a) HDR imaging hardware design; (b) algorithm architecture with color-coded regions corresponding to chapters below; (c) overall processing flow of the framework.
• Figure 2: Channel characteristics in multi-exposure imaging: (a) Representative haze-free scene image with corresponding bright and dark channels; (b) Representative hazy scene image with corresponding bright and dark channels; (c) Grayscale histograms for the four distinct exposure levels of the SVE camera for the scene in (a); (d) Grayscale histograms for the four distinct exposure levels of the SVE camera for the scene in (b).
• Figure 4: Haze perception and segmentation results: (a) Probabilistic haze distribution map with color bar indicating normalized haze density, increasing with haze concentration; (b) scene-adaptive region segmentation ($m$ = 4), with regions labeled by decreasing haze probability (${p_1}$ to ${p_4}$).
• Figure 5: The general pipeline of adaptive pyramid fusion: The illumination and reflection layers are decomposed into Laplacian pyramids, while the weighting maps are decomposed into Gaussian pyramids. The symbol $\times$ denotes the dot product. ${R_1}{\rm{ - }}{R_j}$ represent the layers of the resulting Laplacian pyramid.
• Figure 6: The correlation coefficient maps between features are calculated using (a) PCC and (b) MIC.