Adaptive Object Distance Monitoring in Disaster and Crisis Scenarios via YOLOv9 with Monocular Depth Estimation

Abstract

Situational awareness (SA) is required by the first responders to act swiftly and with accuracy in response to a disaster and crisis situation, where the dynamic nature of the environment, damaged infrastructure and visibility can severely impair the decision-making process. The existing technologies in disaster monitoring fail to conventionally offer real time object localization with accurate distance identification in order to improve the effectiveness of evacuation, rescue and mobilization of resources. To overcome these shortcomings, this paper will suggest an Adaptive Object Distance Monitoring Framework, which combines YOLOv9 to find objects with high accuracy and monocular depth estimation to predict distances in real-time using a single camera stream. The proposed system will identify critical objects (victims, vehicles, and debris) and emergency assets and continuously estimate the spatial proximity of critical objects in the challenging environmental conditions. A refinement and adaptive filtering strategy is introduced to stabilize depth results when there are occlusion, motion blur, and low-light artifacts. The framework allows implementation at low costs on edge and mobile devices and can be used in drones, surveillance cameras, and emergency vehicles. Experimental assessment shows the best detection rates, strong ranging stability, and quick rate of inference as compared to traditional single-task methods. The suggested model can be used to enhance next-generation intelligent disaster resilience systems, where proactive monitoring, quick response coordination, and safer crisis management is possible due to scalable AI-based visual analytics.