Enhancing Urban Search And Rescue Planning: Real-Tıme Integration Of Aerial Object Detection, Rescue Member Observations, And Tracking With Qualitative Spatial Reasoning

Official URL: https://risc01.sabanciuniv.edu/record=b3400657

Locating and tracking potential victims in search and rescue (SAR) missions is a crucial step in real-time monitoring, resource optimization, and mission planning in the aftermath of large-scale disasters. We propose a multi-object tracking paradigm in SAR missions (DODAE) consisting of detection, observation, deduplication, association, and explanation steps. The detection step locates potential victims in aerial images of the disaster region. The observation step collects qualitative spatial relations describing the locations of potential victims relative to landmarks in the disaster area from citizens, volunteers, and rescue team members. The deduplication step filters common observations, and detected false positives and false negatives, via trustworthy observations. The association step links object detections or observations to corresponding track identities. The explanation step describes the locations of potential victims relative to close-by objects. Within DODAE, we introduce two novel methods that utilize qualitative spatial reasoning (QSR): Data Deduplication by Relations (DDR), and Data Association by Relations (DAR). To evaluate these hybrid methods, we build a low frame-rate tracking video dataset to simulate the circumstances of SAR scenarios. We conduct evaluations in four distinct scenarios: with a perfect detector, with detections only, with observations only, and with both detections and observations. The improved accuracy results show the usefulness of QSR for detection and tracking.