This paper presents the updateable probabilistic evacuation modeling (UPEM) technique, which allows sensor observation data to be included in the problem of estimating the state of an evacuating crowd, as the data are obtained. Each individual is modeled as a Newtonian particle which interacts with obstacles, such as walls and other individuals. The UPEM technique estimates not only the general trend of the crowd as a whole, but also the specific states of each of the evacuees in the crowd. Furthermore, an approach to cooperative autonomous searching in crowded urban emergencies is developed using UPEM. A number of simulated searches in emergency evacuations highlight the efficacy of the technique in reducing the time required to detect targets and in increasing the level of safety for human evacuees.