Synergistic Event-SVE Imaging for Quantitative Propellant Combustion Diagnostics

This paper presents a closed-loop Event-SVE imaging system that synergistically combines spatially variant exposure and neuromorphic event cameras to achieve real-time, smoke-resilient, microsecond-resolved 3D diagnostics of high-energy propellant combustion under extreme high dynamic range conditions.

The Gist

Imagine trying to take a clear, high-speed photograph of a campfire while standing inside a thick, swirling fog. You have three major problems:

1. The Fire is Too Bright: The flames are so intense that a normal camera would turn them into a blinding white blob (overexposure).
2. The Fog is Too Thick: The smoke obscures the details, making it hard to see individual sparks.
3. The Sparks Move Too Fast: The embers fly away in microseconds, so fast that a standard camera would capture them as a blurry streak.

This is exactly the challenge scientists face when studying solid rocket fuel (specifically boron-based propellants). They need to see tiny, glowing particles breaking away from the fuel surface to understand how the engine works, but the environment is a chaotic mix of blinding light, heavy smoke, and lightning-fast motion.

This paper introduces a clever solution: a "Super-Team" of two different types of cameras working together to solve this puzzle.

The Two Heroes of the Team

1. The "Smart Eye" (SVE Camera)
Think of this camera as a photographer who can instantly take four photos at once, each with a different filter.

• One photo is taken with a very dark filter to see the blindingly bright fire without it blowing out.
• Another is taken with a clear filter to see the faint details in the thick smoke.
• The Magic: Instead of taking these photos one after another (which would cause motion blur), it does them all in a single split-second snapshot. It then uses a special computer algorithm to "stitch" these four views together into one perfect, high-definition image that shows both the bright fire and the smoky background clearly.

2. The "Speed Demon" (Event Camera)
Think of this camera not as a video recorder, but as a swarm of tiny, hyper-sensitive bugs.

• A normal camera takes a picture 60 times a second. If something moves faster than that, it gets blurry.
• The Event Camera doesn't take pictures. Instead, every single pixel on its sensor acts like a tiny alarm. It only "shouts" (sends a signal) when it sees a change in brightness.
• Because it only reacts to changes, it is incredibly fast (microseconds) and can handle extreme brightness without getting confused. However, it has a flaw: it only sees movement and changes, not the actual picture. It's like hearing a car zoom by but not knowing what color it is.

How They Work Together (The Synergy)

The genius of this paper is how these two cameras help each other, like a detective and a witness.

• The Problem: The Event Camera sees the particles moving fast, but it gets confused by the smoke. The smoke moves, too, and the Event Camera thinks the smoke is a particle. It's like trying to find a specific person in a crowd where everyone is wearing the same mask.
• The Solution: The "Smart Eye" (SVE) takes a clear, high-quality photo and says to the "Speed Demon" (Event Camera): "Hey, look here. The smoke is in these gray areas, but the real glowing particles are in these bright spots."
• The Result: The Event Camera uses the SVE photo as a map. It ignores the "shouts" coming from the smoke areas and focuses only on the particles. This cleans up the data, allowing the system to track the particles perfectly without getting tricked by the fog.

What Did They Measure?

Once the cameras are working together, the system acts like a 3D scanner for fire. By using two Event Cameras looking at the fuel from different angles (like our two eyes), the system can calculate:

1. Separation Height: Exactly how high a particle flies off the fuel surface before it burns out.
2. Particle Size: How big the glowing clumps are.

Why Does This Matter?

Imagine you are designing a rocket engine. If the fuel particles are too big, they might not burn completely, wasting fuel and making the engine unstable. If they are too small, they might burn too fast.

• Before: Scientists had to guess or use slow, blurry cameras that missed the fast action or got blinded by the light.
• Now: This new system gives them a "slow-motion, high-definition, 3D movie" of the fuel burning, even through thick smoke.

The Bottom Line

The researchers built a hybrid camera system that combines the best of both worlds: the clarity of a smart, multi-exposure camera and the lightning speed of a neuromorphic event camera. By letting one camera guide the other, they can finally see the invisible details of how rocket fuel burns, helping engineers build safer, more efficient, and more powerful engines.

It's like giving a detective a flashlight and a super-speed suit at the same time, allowing them to solve a crime scene that was previously too dark and too chaotic to understand.

Technical Summary

1. Problem Statement

High-energy solid propellant combustion (specifically boron-based) presents a unique and challenging diagnostic environment characterized by three coupled factors:

• Extreme High Dynamic Range (HDR): Incandescent particles and flame cores are extremely bright, while surrounding smoke is dark.
• Ultrafast Dynamics: Particle motion and detachment occur on microsecond timescales.
• Dense Smoke Occlusion: Volumetric smoke scatters light, degrading visibility and causing motion blur or saturation in conventional sensors.

Limitations of Existing Techniques:

• High-Speed Cameras: Typically limited to ~60 dB dynamic range, leading to saturation in flame regions and loss of detail in smoke. They also suffer from motion blur at microsecond scales.
• Laser-based Diagnostics (e.g., PIV, PLIF): Suffer from signal attenuation and multiple scattering in optically dense plumes.
• Event Cameras: Offer microsecond resolution and >120 dB dynamic range but lack absolute intensity information, making it difficult to distinguish particle emission from smoke scattering or to perform quantitative size measurements without an intensity reference.

2. Methodology

The authors propose a closed-loop, synergistic measurement system integrating a Spatially Variant Exposure (SVE) camera with a stereo pair of neuromorphic event cameras. The framework consists of four core components:

A. Integrated Optical Platform & Calibration

• Hardware: An SVE camera (capturing 4 interleaved exposures simultaneously via a micro-attenuator array) and two synchronized event cameras (operating asynchronously with microsecond resolution).
• Calibration: Precise spatiotemporal calibration is achieved using Zhang's method for intrinsics and a shared calibration target for extrinsics. Hardware triggers ensure synchronization within 12 µs, allowing sub-pixel alignment between modalities.
• Reconstruction: The SVE raw mosaic is demosaiced into four full-resolution images using gradient-corrected interpolation.

B. Combustion-Specific HDR Fusion (SVE Branch)

To generate a reliable intensity reference, the authors developed a smoke-aware fusion strategy:

1. Smoke-Likelihood Estimation: A probabilistic map $F(x)$ is generated using four features:
   • Brightness Deviation Intensity (stabilizes dark regions).
   • Average Weber Contrast (preserves particle edges).
   • Dark/Bright Channel Contrast (identifies optically thick smoke).
   • Normalized Variance (detects saturation/non-linear regions).
2. Region-wise Retinex Fusion: The scene is segmented based on $F(x)$. A Retinex model decomposes images into illumination and reflection layers. Fusion weights are applied adaptively to suppress smoke in the illumination layer while enhancing particle details in the reflection layer.
3. Output: A calibrated, high-contrast HDR map that serves as an intensity prior for the event cameras.

C. Intensity-Guided Event Processing

The SVE-derived HDR map guides the processing of the event stream to extract particles:

• Particle Classification: Particles are categorized as Actively Combusting (strong emission), Completely Extinguished (weak interface events), or Partially Extinguished (hybrid events).
• Smoke Suppression: Event clusters are filtered using the HDR map as a spatial gate. Only events occurring in regions with sufficient visibility (low smoke likelihood) and matching particle radiative signatures are retained.
• Motion Compensation: For moving particles, a local displacement field is estimated to warp events, stabilizing the particle footprint for accurate contour extraction.

D. Stereo 3D Metrology Pipeline

• 3D Reconstruction: Cleaned 2D contours from both event cameras are triangulated to determine 3D particle centroids.
• Separation Height: Calculated by projecting the particle centroid onto the axial vector of the propellant column.
• Equivalent Size: A pinhole-based spatial scale model converts pixel contours to physical area. The equivalent radius ($r_e$) is derived from the cross-sectional area ($S = \pi r_e^2$), avoiding bounding box overestimation.

3. Key Contributions

1. Hybrid Optical Scheme: A novel integration of SVE and event cameras where the SVE branch provides absolute intensity context to resolve the ambiguity of event data in smoke-filled environments.
2. Combustion-Tailored HDR Algorithm: A specific fusion strategy that decouples particle radiation from volumetric smoke scattering, outperforming general-purpose HDR methods in preserving particle edges while suppressing smoke noise.
3. Microsecond-Resolved 3D Pipeline: A complete workflow from feature extraction to 3D parameter estimation (separation height and equivalent size) with a maximum calibration error of 0.56%.

4. Experimental Results

Experiments were conducted on boron-based propellant combustion:

• HDR Quality: The proposed fusion method achieved lower NIQE (better perceptual quality) and higher NIQMC scores compared to raw SVE inputs and standard PESPD-MEF.
• Feature Preservation: Quantitative analysis showed a 18.4% improvement in Average Gradient (AG) for combusting particle agglomerates compared to the best single exposure, confirming superior edge sharpness.
• Separation Measurement: The system successfully captured the transient detachment of particles. The stereo triangulation yielded a separation height of 15.94 mm with high temporal consistency between left and right views.
• Cross-Modal Consistency: Event-based size estimates showed strong trend agreement with SVE and high-speed camera measurements, though the event camera provided more stable estimates by avoiding motion blur.
• Statistical Analysis: The system resolved multimodal distributions of equivalent particle radii in real-time, identifying "coral-like" aggregate structures that are typically obscured by smoke in conventional imaging.

5. Significance

This work addresses a critical gap in propulsion diagnostics by enabling quantitative, in-situ monitoring of combustion processes under conditions that previously rendered sensors ineffective (simultaneous extreme brightness, speed, and smoke).

• Engineering Impact: Provides a practical route to optimize propellant formulations and suppress combustion instability by accurately measuring particle size distributions and separation dynamics.
• Scientific Advancement: Demonstrates that neuromorphic sensing, when guided by adaptive HDR priors, can overcome the limitations of both traditional frame-based cameras and standalone event cameras, offering a new paradigm for high-speed, high-dynamic-range metrology.