A Target Position Decision System based on Eyewitness Account for Exploring Disaster Environment

The main contribution of the article lies in the realization of the target position decision system for exploring a disaster environment based on eyewitness accounts. This is achieved using the modeling of Bayesian networks to infer locations of interest from eyewitness accounts that are divided into words and the semantic map representation based on image processing. Recently, an exploration of a disaster environment by employing rescue robots before firefighters enter into the dangerous environment has increased in application. This is because getting information in advance, such as a victim's position or a change in environment can protect firefighters in hazardous conditions resulting from smoke, fire or pollutants. However, the gap of information between the symbolic meaning, the information that humans understand, in an eyewitness account and the target coordinates on the map makes it difficult to apply the system for exploring a disaster environment based on robots into the real situation. The proposed system automatically infers the room type of interest using the Bayesian network that is designed based on survey of 70 firefighters and matches it to coordinates from the semantic map. Then, the system distributes the target positions to the robots in order to explore the disaster environment. To verify the performance of the proposed system, we conduct the system verification based on one scenario to make a decision of a target coordinate on the map.