Progressive Ordinates GAN with Centroid Fuzzy Ray-Tracing for Scene Image Detection from Thermal Imaging

Abstract

The security threats from terrorism and illegal migration raised security concerns, insisting the use of thermal cameras in surveillance systems due to their night vision and all-weather capability. However, thermal images suffer from variations in surface roughness, texture, and radiation, hindering accurate object detection. To address this, a Progressive Ordinates GAN with Centroid Fuzzy Ray-Tracing model is proposed for the efficient object detection using thermal images. The existing object detection algorithms struggle in detecting thermal signatures, as they depend only on temperature while neglecting emissivity differences from diverse surface characteristics of thermal images. To overcome this problem, a novel Progressive Ordinates Generative Adversarial Network (POGAN) is introduced, for more accurate detection and characterization of objects with diverse surface characteristics having varying emissivity, mitigating the challenges of inconsistent thermal signatures. Moreover, the shadow casting of objects from thermal image hinders identifying the object boundaries. The existing algorithms struggle to determine the shadow casting as they are not aware of the related parameters such as geometric properties, spatial relationships and the angle of incidence, as they operate on fixed-size image patches or regions. Hence, Centroid Ray-Tracing Fuzzy Clustering (CRFC) is introduced, which effectively acclimates to varying scene complexity by dynamically adjusting cluster centroids, thus enables a better understanding of geometric properties, spatial relationships, and angle of incidence within the scene. The analysis on this work validates that the proposed model achieves better performance with improved accuracy, sensitivity, mean Average Precision (mAP) and recall, with minimized Mean Average Error (MAE).