BROWSE
Detail
Conference Paper
Egocentric Human Pose Estimation for VR Interactions Using RGB-D Data
Cited - time in 
Share 
Share
- Authors
- Seongmin Baek, Yejin Kim
- Issue Date
- 2023-12
- Citation
- International Conference on Internet (ICONI) 2023, pp.294-296
- Language
- English
- Type
- Conference Paper
- Abstract
- Recently, virtual reality (VR) technology has evolved into a metaverse where individuals share information about their real-world activities through head mounted displays (HMDs). For smooth interaction between users, the position and rotation information of the hands and head of a user are recognized by the camera sensor and controller of an HMD while the rest of the body is inferred from the inverse kinematics. However, this can lead to inaccurate pose estimation and reduce the user's immersion in the virtual space. In this paper, we introduce a method to estimate user pose from egocentric (inside-in) data acquired from multiple RGB-D sensors installed on an HMD. In the proposed method, we calibrated multiple RGB images and depth data to the same coordinate system and estimated user pose in 3D using a deep neutral network (DNN). Compared to external sensor-based systems using outside-in data, our method generates accurate user poses in 3D and provides better freedom of movement for VR interactions.
- KSP Keywords
- Camera sensor, Coordinate system, Depth Data, Freedom of movement, Head-mounted display(HMD), Human pose estimation, Outside-in, RGB image, RGB-D data, RGB-D sensors, Real-world