Trinary Relationship Interaction Detection
Fangqun Gao, YunTao Wu, Yanduo Zhang, Tao Lu, Pan Tang, Liwei Wang, Jifeng Han, Wenfeng Wu · 2024
In various specialized operational scenarios, particularly those involving tool usage, the capacity to detect relationships between entities is crucial for understanding complex interactions. Traditional methods, which typically focus on unary or binary relationship detections, often fall short in capturing the subtleties present in such environments. This paper introduces the HTOI (Human Tool Object Interaction) model, a novel and pioneering approach to trinary relationship detection that addresses these limitations. The HTOI model has been designed to excel in complex interaction scenarios, including those characterized by the use of tools, as evidenced by its exceptional performance on our TData dataset. This dataset specifically highlights the nuanced actions of electric utility workers during tool-assisted operations, underscoring the model’s ability to detect intricate interactions with higher accuracy and reliability compared to existing HOI (Human Object Interaction) techniques. Our study aims to showcase the HTOI model’s significant advancement in the field of relationship detection, offering a valuable tool for analyzing and interpreting complex operational interactions involving tools.