Robust human detection system in flood related images with data augmentation

Floods are one of the major natural disasters that are very common in monsoon countries like India. They cause enormous damage, both to properties and to human lives. It is very essential to detect humans in flooded environments, which plays an important role in rescue management, disaster management, flood assessment, etc. Detection in a flooded environment is particularly challenging due to the weather conditions, which affect the efficiency of the detection system. The limited availability of proper datasets with natural weather changes also affects the robustness of human object detection in flooded images. In this paper, data augmentation techniques are introduced in order to mimic the changing weather conditions such as rainy, cloudy, and foggy days during flood times, and a HOG-based Robust Human Object Detection (HOG_based_RHOD) algorithm is proposed and has been demonstrated on the augmented dataset. The proposed HOG_based_RHOD algorithm detects human objects in flood-related images, demonstrating its robustness in a variety of challenging weather conditions.