A privacy-aware Artificial Intelligence framework for crowd monitoring and dispatching using wearable Internet of Things

Large-scale events and mass gatherings create significant challenges for crowd safety, coordination, and privacy-preserving information exchange. This study presents a privacy-aware Artificial Intelligence framework for crowd monitoring and dispatching using wearable Internet of Things (IoT) technologies. The Artificial Intelligence contribution is an agent-based coordination framework that integrates dynamic group management and policy-aware access control, while the engineering application is crowd monitoring and dispatching for large-scale events. Participants, group leaders, and coordination authorities are modeled as autonomous agents within a hierarchical Multi-Agent System (MAS). Dynamic crowd behavior is supported through a Disjoint Set Union (DSU)-based group management mechanism for group formation, splitting, and reintegration under mobility, scheduling, proximity, and health-indicator constraints from simulated wearable sensing. Privacy-aware access control is modeled through Ciphertext-Policy Attribute-Based Encryption (CP-ABE), enabling policy-compliant access to sensitive information. The framework is evaluated through simulation using the Java Agent Development Framework (JADE), involving a 1001-agent scenario, 10 independent repeatability runs, and scalability sensitivity analysis with populations of 2051, 5051, and 10,061 agents. The evaluation also includes a density-based spatial proximity-clustering baseline and crowd-safety-related proxy metrics, including elevated local-density exposure, congestion-pressure proxy, and close-contact risk. Results demonstrate stable coordination behavior, join success rates above 90%, and consistent platform-level latency patterns. Privacy enforcement introduces moderate simulated overhead, with ciphertext size increasing from approximately 992 to 1184 bytes and policy-evaluation cost increasing from 13 to 17 ms as attribute complexity grows. These findings support the feasibility of the proposed framework as a simulation-level architecture for privacy-aware crowd monitoring and dispatching.