Monitoring of motor vehicle exhaust emissions using Gaussian process regression frame interpolation optical flow algorithm

Abstract

In fluid pollutant monitoring, the spatial continuity of pixel motion is disrupted by infrared cameras, primarily due to factors like low frame rate. This disruption impedes the accurate capture of pollutant distribution and evolution, resulting in substantial errors in monitoring outcomes. To address this challenge, we introduce the Gaussian Process Regression Frame Interpolation Optical Flow (GPR-FIOF), aimed at restoring the spatial continuity of pixel motion. Consequently, this facilitates a more precise estimation of fluid pollutant motion. Experimental results from fluid simulations demonstrate that, when compared to conventional algorithms, GPR-FIOF significantly enhances accuracy and stability, improving by 80.30% and 66.39%, respectively. Field experiments employing infrared gas correlation spectroscopy methods revealed improvements in accuracy and stability of emission rate inversion results, with enhancements of 18.24% and 61.77%, respectively. GPR-FIOF effectively mitigates the disruption in spatial continuity, enhancing the accuracy of pollutant gas emission monitoring and bolstering its feasibility for environmental monitoring applications.