Human Victim Detection with Deformable Part Models

Abstract

Algorithms that are highly accurate against object categories that have a sufficiently static shape such as HoG are incapable of recognizing the targets that have a highly articulated figures such as human beings which are the main case of search and rescue missions. In our study, we have implemented a version of DPM that reaches high accuracy and speed when it comes to find these kind of targets in digital images as a part of our goal to bring in the capability of computer vision to our autonomous search and rescue robot.