|
In densely populated regions, the identification of evacuating pedestrians remains a significant challenge, with profound implications for disaster management and public safety. Here, we present a novel approach to identifying evacuating pedestrians under such challenging conditions. Our method integrates attention mechanisms with segmentation techniques to accurately identify individuals amidst the dense smoke and haze of fires. Through extensive experimentation and validation, we demonstrate the robustness and effectiveness of our approach in real-world scenarios. This study not only advances the field of pedestrian recognition but also provides valuable insights for improving emergency response strategies in fire-affected areas. We propose a computer vision approach by augmenting target pedestrian recognition with a mask branch on top of object detection algorithms, utilizing channel attention mechanisms to enhance target pedestrians in the environment while reducing the weighting of the environment in images. We introduce a YOLOv8-Mask model which effectively enhances pedestrian recognition in hazy scenes typical of fire-smoke environments, facilitating efficient rescue operations for trapped individuals. Finally, comparative experiments conducted on self-collected datasets with five established object detection models demonstrate the superiority, robustness, and effectiveness of our proposed model in crowd recognition under such hazy conditions and real-world scenarios. The research outcomes presented herein contribute to assisting rescue efforts in densely populated venues amidst fire-smoke conditions, thereby providing technological support for the development of resilient cities. The model of this study has been made available for public access and can be found on this website https://github.com/xifangfff/YOLOv8-Mask/tree/main
|
