Edge-assisted federated learning framework for smart crowd management

Crowd management is a complex issue that requires a multifaceted approach. It entails employing various strategies and techniques to manage large crowds in urban areas effectively. Crowd management's overall goal is to ensure the safety and security of the crowd and the surrounding community, minimizing disruptions and ensuring the smooth flow of people and traffic. There could be two main challenges in smart crowd management in a smart city environment, i.e., real -time processing of large magnitude of data and privacy preservation. Edge computing devices can be used to process large amounts of data, and privacy concerns can be addressed with the help of federated learning. To address the aforementioned challenges, in this paper, we have proposed an edge-assisted four-layered conceptual framework (i.e., data acquisition, edge computing, cloud computing, and application layers) using federated learning. Practical implications and their potential solutions using the proposed framework are suggested. To fully utilize the benefits of edge-assisted federated learning for crowd management within smart cities, various research challenges are highlighted that need to be addressed. Moreover, a case study in the context of smart cities is presented to highlight the potential benefits of the proposed framework. Simulation results for federated learning-based anomaly detection using three to five local clients are also presented as well to validate the effectiveness of the proposed framework. The effect of label poisoning attacks on FL-based anomaly detection is also discussed.