New AI System Identifies Contactless Human Interactions in Crowded Areas

An innovative AI system, the Interactive Relation Embedding Network (IRE-Net), has been developed to identify and categorize human interactions without physical contact in crowded scenes. This capability could improve public safety monitoring and facilitate social behavior analysis, including during pandemic-related social distancing efforts.

Traditional surveillance systems often focus on direct interactions in less populated settings. In contrast, IRE-Net is designed to operate effectively in complex, real-world environments where multiple individuals are present. Dr. Ruize Han, the project’s corresponding researcher, states, “IRE-Net is built to overcome the shortcomings of current surveillance technologies by analyzing the dynamics of crowded scenes.”

The system uses a combination of appearance and spatial information to analyze interactions. Its core technology involves a structure known as the pairwise-interactive-relation cube, along with a multi-head, multi-task module that predicts the type and context of human interactions.

Preliminary testing indicates that IRE-Net performs better than existing technologies in recognizing non-contact interactions in crowded settings, with improvements noted in reducing false positives and negatives.

IRE-Net also has potential applications in urban planning and crowd management. This technology can provide valuable insights aiding the design and management of public spaces.

Future enhancements for IRE-Net include integrating multi-camera systems to broaden its applicability across diverse environments. “We are planning to expand the system’s capabilities to include a network of cameras, enhancing our analysis of large public areas,” Dr. Han explains.

This research has been published in Frontiers of Computer Science and is a collaborative work between Tianjin University and University of South Carolina. The complete study is accessible via DOI: 10.1007/s11704-023-2418-0.