CORE: A dataset of critical objects for response to emergency

Abstract

Robotic first responders have potential to significantly improve rescue efficiency and safety in search and rescue missions. To operate intelligently, a robot requires the capability to recognize critical objects in a disaster environment, in order to effectively locate victims and/or prevent secondary disasters. In this report, we introduce a novel dataset of Critical Objects for Response to Emergency (CORE) to facilitate future design of object detection systems for search and rescue missions. We also implement an object detection approach, using object proposals, deep features, and classifiers, to recognize objects in the CORE dataset. An average accuracy of 94.6% is achieved.