A holistic approach to human presence detection on man- portable military ground robots

Object detection is a well-explored problem in computer vision. Although object detection algorithms have improved greatly in terms of speed and accuracy, many of these algorithms cannot be used directly on man-portable ground robots in-theatre due to poor detection rates and/or runtime efficiency when exposed to the low-quality camera data and processing limitations of these robots. In order to facilitate the introduction of near-term practical object detection capabilities to these robots, this thesis argues that a "holistic" approach should be taken which views the robot as a system with its vision being only a small part of its senses. This thesis analyzes many different aspects of man-portable ground robots from the history of their use to technological and philosophical obstacles to deploying these robots with autonomous behaviors. To demonstrate this "holistic" approach, this thesis introduces a human presence detection (HPD) method which utilizes the robot's ability to physically maneuver itself which will hopefully lead to better detection rates. This is simulated using difficult multi-scale testing data designed to expose the strengths and weaknesses of the HPD method in this thesis and to pave the way for future development. It is hoped that further integration of the robot's behaviors, the use of a multitude of different sensors, and improved interaction with human soldiers will lead to object detection capabilities (such as HPD) which will be deployable on man -portable ground robots currently used by soldiers in- theater