Human Target Recognition in Dense Smoke Environments based on Multimodal Input and Image Smoke Removal Fusion Algorithm

Modern fire protection systems incorporate various sensors such as infrared and thermal imaging technologies. However, in actual fire rescue scenarios, accurate detection of human targets in smoke-filled environments remains significantly challenging due to the attenuation of visible and infrared light by dense smoke, complex ambient thermal fields, and interference from obstructing objects. To address this issue, this paper proposes a multimodal smoke-removal fusion algorithm for the precise identification of human targets in smoke-obscured environments. The process begins with spatiotemporal alignment of raw visible- light and infrared video streams to synchronize image frames. Subsequently, multiple image processing techniques–including Contrast-Limited Adaptive Histogram Equalization (CLAHE) and median filter denoising–are applied to mitigate smoke occlusion effects. Finally, human detection is performed using YOLOv8. The video annotation results demonstrate that the proposed algorithm exhibits commendable smoke removal performance. Specifically, the YOLOv8m and YOLOv8n models achieved precision rates of 99% and 97.7%, respectively, in human target detection under dense smoke conditions. This indicates substantial potential for applications in fire rescue operations, security surveillance systems, and industrial safety inspections.