Accelerated Testing of an Image Classification Algorithm for Emergency Cardiac Incident Response

Rahul Gomes


If cardiac arrest occurs, the victim’s chances of survival decrease if he or she is not attended to immediately. Problematically, however, not everyone will be near a defibrillator with a trained user when experiencing cardiac arrest. Automated external defibrillators (AEDs) represent a partial solution; however, they still require a victim to be attended and near the AED. An autonomous unmanned aerial vehicle has been proposed to deliver and apply AEDs within buildings, stadiums and other constrained areas. The system is equipped with an image classification algorithm that is used to identify an appropriate location and protocol for AED attachment, based on body parts identification. The algorithm collects and processes 3D point cloud data. The testing of this system is critical to ensure that it connects the AED to only appropriate body locations. This paper presents an accelerated testing regime designed to validate the functionality of the placement algorithm and system.