Artificial Intelligence (AI) has been widely used to facilitate disaster response. By connecting cameras to AI software, it can help determine the number of firefighters and apparatus, enhancing efficiency on the fireground. However, we must overcome several challenges to effectively utilize AI in firefighting. One challenge is improving the brightness and resolution of pictures and videos taken at fire scenes. This study examines the impacts of two image enhancement methods, Contrast-Limited Adaptive Histogram Equalization (CLAHE) and Zero-reference Deep Curve Estimation (Zero-DCE), on the accuracy of the AI-based object detector trained using images taken on various fire scenes. The results indicate that, after augmenting the training data with image enhancement techniques, the detector can accurately identify firefighters with a precision of 0.827 and firetrucks with a precision of 0.945. Enhancing the dataset's variety through these techniques improves the model's generalizability, provided that the test images are also enhanced to augment visual quality. Specifically, applying CLAHE during training increased the mean average precision (mAP) value by 8% and the recall by 7% from the baseline. Meanwhile, the integration of Zero-DCE demonstrated particular efficacy in recognizing firetrucks in low-light conditions, achieving the highest precision value of 0.945 among all the cases considered. This paper will benefit future applications of AI in fireground operations. Additionally, we provide directions for future researchers to advance AI recognition research in facilitating disaster response activities and fireground operations.
Keywords: Artificial intelligence (AI); data argumentation; firefighter safety; object detection; occupational safety and Health.