The Active Shooter Tracking & Evacuation Routing for Survival (ASTERS) project was a collaborative effort between University of Tennessee, Knoxville and Iowa State University that brought together various expertise such as computer vision and machine learning, planning and optimization, human-machine interactions, education, and social sciences. The overarching goal of the project was to develop a deployable technological solution to mitigate the impact of active shooter incidents in K-12 schools and reduce resulting injuries.

To achieve this goal, the ASTERS team built a cyber-physical system (CPS) framework that leverages multiple cameras deployed in a school building to localize, identify, and track a shooter using computer vision and machine learning techniques. If a shooter is identified, the route optimization module estimates the safest evacuation routes for different groups of students and school employees in real time. Information about the shooter is communicated to appropriate school officials and law enforcement personnel automatically, and personalized evacuation guidance is provided to people under threat via the ASTERS communication module.  

The training process of a camera-based shooter detection model required a large amount of data under widely different conditions in terms of locations, lighting, types of shooters, and clothing. However, publicly available data on shooters in a security context remains quite limited. Therefore, the ASTERS team explored the use of computer graphics game engines such as Unreal Engine (UE) to build a large public data set. The team developed machine learning algorithms that used this large synthetic data set augmented with a small amount of real data to build highly accurate shooter detection models that generalize well to different active shooter situations. The shooter detection and tracking system has been implemented to run on off-the-shelf, low-cost edge hardware including Raspberry Pi and Jetson Nano.

In the ASTERS route optimization framework, the team developed new mathematical models for the shooter’s movement inside a building. Using this shooter movement model, the team designed a machine learning algorithm that computes the optimal policy for the evacuees by computing the risk versus reward tradeoff of the short-term objective of getting to a safer room, and the long-term objective of getting out of the building safely. In simulation, this algorithm was shown to reduce casualties by 56% and the evacuee’s time spent in the shooter’s line of sight by 52%. The framework also addresses the bottlenecks and crowding problems that can significantly reduce evacuation efficiency. The team also investigated the effect of dynamic opening and closing of programmable automatic doors to control the shooter’s movement. The doors should be strategically placed in important parts of the building to actively steer the shooter away from large crowds, while simultaneously ensuring safety for all evacuees. The ASTERS automatic door actuation framework can significantly reduce evacuee casualties in certain scenarios. The results also helped quantify the abstract ethical questions about equitable risk distribution in the evacuation algorithm.

To evaluate the feasibility of effective personalized communication of evacuation instructions to members of a school community, the team developed a Unity-based school shooter simulation in which research study participants moved through the school to a classroom and then evacuated as quickly as possible when a shooter alert was announced within the school. Participants received instructions such as smart directive Exit signs that changed according to the location of the shooter and smart overhead audio speakers that issued customized verbal evacuation instructions depending on the location of the speaker. Results showed that the audio cues were more effective in general than the visual cues. To measure the effectiveness of an evacuation protocol, a student-led research project inspired by ASTERS created the EASE (Evacuation with Acceptable Simplicity in Emergencies) score that leverages both previous literature on fire and shooter evacuation and results from ASTERS surveys. The EASE agent-based simulation software is available as an open-source repository for use by future researchers.

The ASTERS team also engaged the end-user community such as patrol officers and school leaders through interviews to discuss the relative merits of various choices of interfaces for disseminating mission-critical information to 9-1-1 mission control and real-time survival strategies to civilians. The school leaders also helped to understand the feasibility, potential cost, and real-world application of the ASTERS system. With their help, the team developed an open-source online curriculum module to train other school leaders in developing safety plans and relationships with community members.

The ASTERS project partially supported graduate studies of 13 students (40% women or underrepresented minority) including 10 PhD students. An undergraduate CPS minor at Iowa State was launched in 2021 and the ASTERS project directly supported the development and growth of this minor.

Last Modified: 01/28/2024
Modified by: Soumik Sarkar