A Dual Asynchronous--Synchronous Relation Graph Method for Sensor-Based Group Activity Recognition

doi:10.1109/jiot.2026.3696413

Return

A Dual Asynchronous--Synchronous Relation Graph Method for Sensor-Based Group Activity Recognition

2026-05-25

PRE

DOI：10.1109/jiot.2026.3696413

Original

Original request for help

Save

Abstract

En 中文

The fast advancement of the Internet of Things (IoT) has laid a foundation for group activity recognition (GAR) in smart environments, which aims to understand complex activities composed of multiple individual actions. However, existing methods primarily focus on recognizing synchronous actions and exhibit significant limitations in handling asynchronous actions, which are prevalent in dynamic real-world IoT scenarios. In asynchronous actions, temporal-delayed individuals perform or respond to actions with temporal lags that are not synchronized with those of real-time individuals, forming complex temporal-delayed relations. To address this issue, this article proposes a dual asynchronous–synchronous relation graph (DASRG) method for sensor-based GAR. By introducing an asynchronous relation graph (ARG), the temporal-delayed relations between the temporal-delayed individuals and real-time individuals are captured, and the features of temporal-delayed individuals are synchronized with those of the real-time individuals through feature shifting. Meanwhile, a synchronous relation graph (SRG) is constructed to enhance the modeling of interactions and action differences between individuals by integrating spatial interaction relation (SIR) features and residual relation features (RRFs), thereby reducing the interference of noncritical individuals on recognition results. To achieve the coupling between asynchronous and SRGs, a graph aggregation (GA) module based on the multihead attention mechanism is designed. By leveraging weighted feedback from attention heads, this module optimizes node information updating and enables the mutual feedback of information between the two graphs in the spatial domain, improving the model’s ability to handle asynchronous actions and mitigate interference from noncritical individuals. Extensive experimental results further demonstrate the superiority of the proposed method in complex scenarios involving asynchronous actions and interfering individuals.

Keywords:

Asynchronous relation graph (ARG)

group activity recognition (GAR)

Internet of Things (IoT)

sensor data

synchronous relation graph (SRG)