Modeling Touch Point Distribution with Rotational Dual Gaussian Model
summary
Proposes the Rotational Dual Gaussian (RDG) model that orients the bivariate Gaussian major axis along finger movement and uses projected width and height (via eigendecomposition) instead of nominal dimensions; evaluated on Ko et al.’s dataset (63 target sizes, 16 directions) with WAIC model selection and shows better fit, reduces RMSE from 8.49% to 4.95% for error-rate prediction. Implements a statistical decoder and tests on typing: 29000 labeled touch points from 14 smartwatch users (accuracy up for 11/14) and improves 6/10 smartphone users.
AI-written summary · about
chapters (11)
- 0:00Intro
- 0:20Dual Gaussian Model
- 1:20Rotational Dual Gaussian (RDG) Model
- 2:03Defining Width and Height
- 2:52Evaluation on 2D Target Acquisition Tasks
- 3:29Evaluation: Predicted Distribution of Touch Points
- 4:08Evaluation: Predicted x-y Correlation
- 4:42Evaluation on Smartwatch-based Typing
- 5:12Evaluation on Azenkot and Zhai's dataset [1]
- 5:31Applications of the RDG Model
- 6:05Conclusion
related talks
39mLocalization and Nonmetric Multidimensional Scaling via Paired Comparisons
21mRegression by Eye: Estimating Trends in Bivariate Visualizations
09mTangibleData: Interactive Data Visualization with Mid-Air Haptics
19mInvestigating the Use of Spatial Interaction for 3D Data Visualization on Mobile Devices
38mWhere are we looking? Predicting human gaze using deep networks.
33mDynamic Object Gaze Tracking with OpenCV
10mVirtual Object Categorisation Methods: Towards a Richer Understanding of Object Grasping for ...
05mVisualizing distributions and uncertainty using ggdist