Human Victim Detection and Stereo - based Terrain Traversability Analysis for Behavior - Based Robot Navigation

4.1 Introduction Crisis management teams (e.g. fire and rescue services, anti-terrorist units ...) are often confronted with dramatic situations where critical decisions have to be made within hard time constraints. In these circumstances, a complete overview of the crisis site is necessary to take correct decisions. However, obtaining such a complete overview of a complex site is not possible in real-life situations when the crisis management teams are confronted with large and complex unknown incident sites. In these situations, the crisis management teams typically concentrate their effort on a primary incident location (e.g. a building on fire, a crashed airplane ...) and only after some time (depending on the manpower and the severity of the incident), they turn their attention towards the larger surroundings, e.g. searching for victims scattered around the incident site. A mobile robotic agent could aid in these circumstances, gaining valuable time by monitoring the area around the primary incident site while the crisis management teams perform their work. However, as the human crisis management teams are in general already overloaded with work and information in any medium or large scale crisis situation, it is essential that such a robotic agent-to be useful-does not require extensive human control (hence it should be semi-autonomous) and it should only report critical information back to the crisis management control center. 2 In the framework of the ViewFinder project, such an outdoor mobile robotic platform was developed. This semi-autonomous agent, shown on Figure 4.1, is equipped with a differential GPS system for accurate geo-registered positioning, and a stereo vision system. The design requirements for such a robotic crisis management system give rise to three main problems which need to be solved for the successful development and deployment of such a mobile robot: 1. How can the robot automatically detect human victims, even in difficult outdoor illumination conditions? 2. How can the robot, which needs to navigate autonomously in a totally unstructured environment, auto-determine the suitability of the surrounding terrain for traversal? 3. How can all robot capabilities be combined in a comprehensive and modular framework, such that the robot can handle a high-level task (searching for human victims) with minimal input from human operators, by navigating in a complex, dynamic and environment, while avoiding potentially hazardous obstacles? In response to the first question, we present an approach to achieve robust victim detection in difficult outdoor conditions, by going out from the …