Human engagement intention intensity recognition method based on two states fusion fuzzy inference system

Natural human–robot interaction requires social robots to have human-like perception of engagement intention. In the multi-person interaction scenario, it’s a vital task for social robots to rationally select the main interaction object. Existing studies mostly focus on analyzing whether a person has engagement intention before interaction. However, this qualitative analysis of engagement intention is only applicable in single-person interaction scenarios. When multiple people have the intention to engage with the robot, the robot needs to quantitatively analyze the engagement intention intensity (EII) of all people to make a reasonable interaction decision. In addition, for EII recognition, it is an ideal state that social robots can imitate human social thinking as much as possible. For these purposes, a method that can efficiently recognize the EII by fusing transient features and temporal features is proposed. First, the 3D pose extractor is used to extract the 3D skeleton information which can calculate the transient features including linear distance and body orientation. Second, an improved ConvLSTM network is proposed to effectively identify pedestrian motion states which can reflect temporal information. Finally, based on the proposed two states fusion fuzzy inference system (TSFFIS), the EII can be judged by the three features which are linear distance, body orientation and motion states. Comparative experiments show that our method can effectively identify the EII of different pedestrians relative to the robot. Compared with existing methods, the EII recognition method based on TSFFIS has better performance.