Knowledge-Based Multimodal Intention Active Perception Algorithm Research
Jie Hong Yuan, Zhiquan Feng, Jinglan Tian, Xue Fan, Qingbei Guo · 2020
With the aging of the population, more and more old people need to live alone at home. In order to improve the robot's ability to understand the user's intention in the home environment, this paper proposes a knowledge-based multimodal intention active perception algorithm. The robot actively perceives the user's intention to interact before the user issues interactive commands. Finally, the algorithm was verified on Pepper robot. On the basis of knowledge, this algorithm can actively understand the user's intention by integrating multimodal information. First of all, visual intention inference and auditory intention inference are obtained by combining database and expert knowledge base. Then, the fusion results are obtained by using the multimodal intention active perception algorithm. The robot initiatively initiates interactions to request user feedback and perform corresponding actions. Experimental results show that the proposed multimode algorithm is superior to the single-mode algorithm. It improves the efficiency of human-computer interaction and reduces the burden of users, and is well received by test users.