A Human-Computer Interaction Method for Intelligent Pan-Tilt Based on the Combination of Head-Eye Posture
Tian Gao, Limei Song, Ang Li, Haomiao You, Yinghao Lu · 2023
This paper presents the design of an intelligent pan-tilt human-computer interaction system that can be utilized for distance teaching through a combination of head-eye posture. The system obtains real-time head-eye posture information of the remote teacher using an ordinary camera on the teacher side. The data is then sent to the student side to drive the pan-tilt in real-time to track and rotate to the corresponding field of view. The real-time status value is returned to the teacher side computer to achieve remote pan-tilt human-computer interaction. The proposed method employs the ERT (Ensemble of Regression Trees) face alignment method of cascade regression, which is a gradient boosting-based regression tree method used to extract feature points from real-time frames. The coordinate information of feature points is stored to the set, and then the face contour area is traversed and filtered to select the target object and perform head pose estimation. During the experiments, users provided positive feedback on the system's real-time performance and accuracy. Experimental results demonstrate that the average recognition rate of the head-eye pose combination algorithm can reach 98.33%, and the average accuracy of the pan-tilt interactive experiment can reach 94.35%. The proposed method overcomes the limitations of remote space vision, allowing users to obtain multi-dimensional full-view video according to their posture changes. This provides great convenience for distance learning, and the system has many potential applications in the military, medicine, and transportation fields.