At Lund University in Sweden, Drs. Sebastian Nilsson and David Sanned (now at Saab) are working to measure and visualize fast volumetric processes in plasma and flame applications.

Their research focuses on emission tomography. Synchronized cameras record the light emitted by a volume from several angles simultaneously. A custom in-house tomography algorithm then reconstructs the 3D volume over time, producing a 4D sequence that can be examined frame by frame for both plasma and flame diagnostics.

“Emergent’s cameras are cost-effective, powerful and mobile. The eSDK was well documented, and your application engineering support has been tremendous. Overall, we’re very happy with our camera and platform choice.”
— Sebastian Nilsson, Researcher, Lund University

Nilsson and Sanned arrived at the project from complementary backgrounds. Sanned’s doctoral work focused on 3D tomography techniques for combustion and flame emission, while Nilsson brought experience in plasma research, laser diagnostics, and thermal processing.

Their collaboration began with a shared challenge: planar laser diagnostics could identify species and properties within reacting flows, but it was hard to place those 2D slices within a moving 3D flame front or plasma arc.

In earlier work, the team used cameras running at 50 frames per second. That proved the concept, but the frame rate was too low to follow fast events as they developed in flames and plasmas.

As Nilsson explained in our discussion, the missing gap was that they could “probe or view the final state” but could not understand how the plasma arrived there.

For high-speed emission tomography, that limitation is critical. Unlike a stationary object in a medical CT scan, flames and plasmas fluctuate constantly. The volume cannot be scanned sequentially from one angle after another; all camera views must be captured at the exact same instant, and at high enough frame rates to track the evolution of the event.

How Emission Tomography Works

Emission tomography works in reverse compared to traditional X-ray imaging. Instead of passing external radiation through an object to measure absorption, the cameras capture the light emitted by the process itself—such as flame chemiluminescence or plasma emissions. Each camera records a 2D projection from its line of sight, and a custom in-house algorithm reconstructs the original 3D emitting volume from these synchronized views.

The scientific advantage is that the technique is not limited to surface appearance. The reconstruction provides a full-field view, allowing researchers to study the internal volume of the emitting object, not just its exterior.

Figure 2 | Principles of 4D optical emission tomography:

(A) A self-emitting volumetric source (turbulent flame chemiluminescence or non-thermal plasma) radiates omnidirectionally. Synchronized camera arrays capture line-of-sight line integrals without external illumination or beam attenuation.

(B) 3D spatial discretization into a regular voxel grid f(x,y,z). Sparse forward projection weight matrices W_ij model the geometric contribution of each voxel to detector pixel p_i. Iterative Krylov subspace solvers (LSQR) regularized by spatial smoothness constraints then invert the system to reconstruct the internal 3D emission density.

Achieving 4D Emission Tomography with 100GigE Cameras

To move to high-speed 4D emission tomography, the Lund University team built a six-camera array using Emergent ZENITH HZ-2000-G monochrome cameras. The cameras surround the experiment, each covering a different angle. Their compact form factor makes them practical to reposition across different experimental campaigns and test rigs where space and optical access are tight. Custom in-house software then processes the synchronized streams into volumetric reconstructions that can be rotated, paused, and analyzed in detail.

Figure 3 | Six-camera 100GigE emission tomography setup.

Six Emergent Vision Technologies ZENITH HZ-2000-G monochrome cameras equipped with optical filters positioned circumferentially around a central burner flame on a vibration-isolated optical table. Geometric calibration uses a precision multi-planar dot target. Sub-microsecond exposure synchronization across all six camera heads is maintained via Precision Time Protocol (PTP / IEEE 1588) over 100GigE QSFP28 optical fiber links.

The HZ-2000-G’s Gpixel GSPRINT4502 CMOS sensor delivers 2048 x 1216 resolution at up to 3,463 frames per second in 8-bit mode. The Lund setup typically runs with 2×2 binning (1024 x 608) at 2,000 fps, and reaches up to 10,000 fps with region-of-interest (ROI) windowing to resolve rapid flame instabilities and plasma transients. The cameras stream over 100GigE QSFP28 interfaces. David Sanned notes that while four cameras are roughly the minimum needed for mathematical convergence, six provide solid reconstruction quality. Adding more cameras up to 8–10 would improve spatial fidelity, but with diminishing returns.

Precision Time Protocol (PTP) was used to synchronize recording across the six cameras. Every camera shares the same clock over Ethernet. When the tomographic reconstructions are rendered, phenomena can be observed at the same time from multiple perspectives.

To ensure that the supporting hardware met the performance requirements of the application, the Lund University team and Emergent used a consultative approach. This helped to make effective cost-benefit choices.

A workstation was built consisting of an AMD Threadripper CPU, large memory capacity, and three Broadcom 100GigE network cards. An Intel Arc Pro GPU was selected in part for its available VRAM, which could handle the computationally intensive tomographic reconstruction. This setup could meet the team’s price-to-performance requirements.

On the software side, Nilsson and Sanned built their own acquisition interface and processing pipeline using Emergent’s eSDK Pro. That flexibility was essential: they needed a platform that integrated into a custom scientific workflow rather than a closed commercial package.

Figure 4 | High-throughput acquisition and reconstruction pipeline.

(A) Custom TomoView multi-camera synchronized acquisition and replay interface developed using Emergent eSDK Pro, displaying simultaneous live 2,000 fps streams from all six ZENITH cameras with zero packet drop and sub-microsecond PTP synchronization lock.

(B) Workstation architecture featuring an AMD Ryzen Threadripper CPU, 256 GB system RAM, Intel Arc Pro GPU for accelerated inverse solver computation, and three Broadcom dual-port 100GigE QSFP28 NICs receiving zero-copy RDMA / RoCEv2 streams directly into system RAM.

Sebastian and David used eSDK Pro to capture, process, transfer and store imaging data. Sebastian described the Emergent software development kit as “well documented and efficient,” and said Emergent’s support in choosing hardware was very helpful.

Choosing the Broadcom NICs meant also being able to deploy RDMA / RoCEv2, which placed their imaging data directly into application memory—critical for capturing data from the six 100GigE cameras without CPU bottlenecks or frame drops. The Broadcom NICs are also a cost-effective choice versus more expensive alternatives.

Visualizing Dynamic Phenomena in 4D

Figure 5 | Multi-angle projections and reconstructed 3D emission density.

(A) Synchronized 2D projection frames captured simultaneously by the six perimeter ZENITH cameras at 2,000 fps (1024 × 608 resolution) showing line-of-sight flame chemiluminescence.

(B) Corresponding 3D tomographic volume reconstructed at the identical time instant using iterative LSQR with space carving, exported as an OpenVDB voxel grid and volumetrically rendered in Blender to reveal internal reaction zone topology.

Video 1 | 4D Volumetric Visualization of Turbulent Flame Front

High-resolution volumetric render (OpenVDB / Blender) showing full 3D spatial rotation and temporal evolution of flame chemiluminescence reconstructed from the six-camera ZENITH array at 2,000 fps.

Video 2 | 4D Tomographic Reconstruction of Gliding Arc Plasma Discharge

3D volumetric tracking of transient non-thermal plasma filament motion, elongation, and restructuring recorded at up to 10,000 fps.

Video 3 | High-Speed Direct Optical Recording of Gliding Arc Discharge

Direct physical high-speed camera recording illustrating cyclic restrike and gliding behavior of atmospheric plasma filaments between converging electrodes.

To reconstruct the 3D volume from the 2D camera views, the researchers developed a custom tomographic algorithm in Julia. It uses camera calibration and space carving to determine the physical boundary of the emitting region, then iteratively solves the inverse problem to map emission values across the volume. The reconstructed voxel arrays are exported to OpenVDB format and rendered in Blender, taking about one minute per frame. At capture speeds up to 10,000 frames per second, the resulting 4D reconstructions capture transient flame and plasma dynamics on microsecond timescales.

For Nilsson and Sanned, the goal extends beyond their own lab. They want to lower the barrier to multi-angle high-speed volumetric imaging for researchers who lack the budget for traditional ultra-high-speed camera rigs.

By pairing modern 100GigE machine vision cameras with custom reconstruction software, the Lund team has built an accessible route to 4D emission tomography for complex industrial and physical systems.

Figure 6 | Multimodal diagnostic integration

Six-camera high-speed emission tomography rig coupled with planar laser-induced fluorescence (PLIF) and spectroscopy optics on an optical rail system, enabling simultaneous 3D structural tracking and localized species/temperature measurements.

The setup gives researchers a clearer view of fast-evolving systems. By adjusting parameters such as fuel mix, gas composition, electrical power, and pulse frequency, they can track how volumetric emission structures shift over time. This is useful across combustion, plasma pyrolysis, spray dynamics, metal powder production, and thermal coatings.

Combining 4D emission tomography with laser diagnostics and spectroscopy adds another layer of information. Planar lasers and spectrometers excel at measuring localized temperatures, chemical species, and radicals, but they lack wide 3D context. Tomography supplies the missing piece by tracking the full 3D geometry and motion of the flame or arc in real time. Pairing these measurements lets researchers tie chemical kinetics and local temperatures directly to the surrounding volumetric motion.

Further Articles for Reading