“Real-time Avatar Motion Synthesis by Replacing Low Confidence Joint Poses” by Tanaka, Mori, Toyama and Shoji
Conference:
Type:
Entry Number: 85
Title:
- Real-time Avatar Motion Synthesis by Replacing Low Confidence Joint Poses
Presenter(s)/Author(s):
Abstract:
In recent years, the availability of low-cost depth sensors has facilitated motion capture applications for personal use. Human motion can be captured to control an animated avatar, with no requirement to wear the dedicated sensors. However, users must contend with avatar pose errors resulting from user pose recognition failures, which can be caused by usage environment problems or measurement errors. For these problems, the accurately capture system using a single depth camera [Wei et al. 2012] have been proposed. In this paper, we propose a simple method for synthesizing seemingly natural avatar motion based on the user’s body movements, including user pose recognition failures (Figure 1). First, we calculate the degree of confidence for each joint’s pose parameters that are captured using a depth sensor. Next, the low confidence joint poses are replaced with a similar pose that is calculated based on high confidence joint poses. In addition, the joints that are not detected are complemented with a calculated similar pose.
References:
- Wei, X., Zhang, P., and Chai, J. 2012. Accurate realtime full-body motion capture using a single depth camera. ACM Trans. Graph. 31, 6 (Nov.), 188:1–188:12.
