Concentration-aware real-time 3D reconstruction of leaking gas clouds based on two-view bandpass optical gas imaging

Gas diffusion trend tracking and concentration perception are critical challenges in chemical safety. Existing methods suffer from different limitations such as restricted sensing range, sluggish response, high costs, and weak localization. In sight of this, we endeavor on a scarcely explored task known as real-time reconstruction of gas clouds based on two-view signals, aiming to advance dynamic 3D gas cloud visualization and concentration characterization. Based on cost-efficient bandpass OGI systems, we decompose the task into two main components: (i) establishing the relationship between images and gas column density, and (ii) reconstructing 3D gas clouds from two views. Initially, we revisit the imaging mechanism and establish a comprehensive spectral signal transmission model, introducing a physics-driven method for column density inversion. Subsequently, inspired by the Gaussian dispersion model and the system's observational mode, we introduce a gas cloud real-time reconstruction network that facilitates the reconstruction of gas distribution through angle encoding and spatiotemporal feature fusion. Experimental results demonstrate that the inversion method achieves low relative errors, and the reconstruction algorithm effectively models geometrically irregular and ever-changing gas at the video level. Noteworthily, real-world experiments validate that the proposed pipeline enables rapid perception of gas diffusion trends and 3D concentration distributions, providing a new avenue for real-time gas leak monitoring.