Publication

Abstract : The design and development of a rover system for battlefield casualty assessment and decision support is presented. The proposed system integrates computer vision, machine learning, and environmental sensing to assess human casualties in hazardous environments, providing a structured situational awareness report for remote rescue decision support. The rover uses a Raspberry Pi with a Pi Camera for vision processing, while an ESP32 microcontroller controls rover locomotion and the onboard environmental sensors such as temperature, fire, and gas detectors. Human detection and landmark extraction are used with MediaPipe Pose in the vision pipeline. A Random Forest classifier trained on the publicly available LE2I Fall Detection Dataset achieved more than 90% accuracy using six biomechanical features extracted from pose landmarks to identify falls. The temporal stability of a potential casualty is analyzed within a 60-second observation window to assess the mobility of the subjects. The results are summarized in one concise report that contains details such as human detection status, fall classification confidence, posture state, mobility test results, and environmental hazard readings. It is not meant to replace human decision-making, which remains the responsibility of the operator, but to augment it. The report is presented to the operator, who decides whether to deploy a rescue team. Experimental results show that this pipeline can be deployed on edge devices with limited resources and real-time inference can be performed on the Raspberry Pi platform.