Researchers at the Hefei Institutes of Physical Science, part of the Chinese Academy of Sciences, have developed a fast three-dimensional imaging system that detects and reconstructs the shape and spread of gas leaks in real time, moving beyond traditional two-dimensional methods that mostly show only a projection of a gas plume. Tech Xplore tells that the new system combines multispectral imaging hardware with advanced algorithms and machine learning to deliver accurate 3D data on gas volume, distribution, and source location within a few hundred milliseconds, a capability that could improve emergency response, industrial safety checks, and environmental monitoring.
The imaging setup integrates infrared detectors, lenses, and motorized components, and uses a trained YOLOv10 model to identify gas plumes at more than 25 frames per second. Once gas is detected, a mathematical technique called a non-axisymmetric inverse Abel transform reconstructs the plume’s three-dimensional form in less than 200 milliseconds with high fidelity. Initial simulations reported strong image quality metrics, indicating accurate volume and spatial mapping down to fine details.
For larger gas clouds spread over wide areas, the team built a more powerful imager and coupled it with a deep learning-based 3D reconstruction network that leverages octree data structures. This network builds a detailed model of the gas cloud from coarse to refined resolution using limited computational resources. Field tests confirmed that the system can effectively capture the shape, position, and spread of real leaks, making it suitable for both industrial sites and broader environmental applications where quick spatial awareness is essential.
Detecting toxic or flammable gases quickly and knowing their exact location and diffusion patterns are vital for preventing fires, explosions, and ecological harm. This work gives researchers and safety engineers a tool that not only spots leaks but also visualizes them in three dimensions fast enough to guide decisions during unfolding incidents, potentially reducing risk to people, infrastructure, and surrounding ecosystems.