Modeling Touch Point Distribution with Rotational Dual Gaussian Model

Title of the Paper

Modeling Touch Point Distribution with Rotational Dual Gaussian Model

Paper Information

• Domain: Human-Computer Interaction (HCI) and Touch Interface Design
• Keywords: touch input, modeling, pointing distribution, rotational dual Gaussian model, touchscreen interface design, target acquisition, movement direction, keyboard decoding, touch accuracy, user interface

Research Background and Problem

• Identified Issues or Challenges:
  • Existing dual Gaussian models fail to account for the influence of finger movement direction on touch point distribution, limiting predictive adaptability.
  • Finger movement direction elongates the touch point distribution along the movement axis, a phenomenon not adequately captured by current models.
• Significance:
  • Accurately predicting touch point distribution is critical for designing touchscreen interfaces, target acquisition tasks, and soft keyboard decoding.
  • Incorporating finger movement direction can improve model prediction accuracy, thereby optimizing user experience and input interface design.
• Motivation and Related Work:
  • Dual Gaussian models have been widely applied in touchscreen design and soft keyboard decoding.
  • Previous studies, such as 1D and 2D target pointing prediction models and experiments on the effect of movement direction on touch point distribution, have validated the importance of movement direction.
  • This study aims to extend the dual Gaussian model by integrating finger movement direction into the modeling process.

Solution

• Proposed Method or Solution:
  • Introduced the Rotational Dual Gaussian Model, which incorporates finger movement direction (θ) into touch point distribution prediction.
  • In the model, the major axis of the predicted ellipse aligns with the finger movement direction, while the minor axis is perpendicular to it.
  • Introduced "projected width" ((W_p)) and "projected height" ((H_p)) as constraint conditions.
• Innovations:
  • Compared to traditional dual Gaussian models, the Rotational Dual Gaussian Model captures directional changes in touch point distribution.
  • By considering projected dimensions and finger movement angles, it more accurately reflects actual touch point shapes.
• Implementation Steps and Key Techniques:
  1. Established a new covariance matrix formulation, mathematically linking finger movement direction θ with touch point distribution.
  2. Conducted model parameter estimation using Bayesian methods to calculate parameters (a, b, c, d).
  3. Compared three alternative approaches: projected dimensions ((W_p), (H_p)), apparent dimensions ((W_a), (H_a)), and nominal dimensions ((W_n), (H_n)) to evaluate model performance.
  4. Validated the model's predictive capabilities on three datasets, including target acquisition tasks and typing tasks in work environments.

Research Outcomes

• Specific Results:
  • Compared to the original dual Gaussian model, the Rotational Dual Gaussian Model significantly reduced prediction errors.
  • In target acquisition tasks, RMSE error rate decreased from 8.49% in the original model to 4.95%.
  • Using the rotational model as the spatial model for soft keyboard decoding significantly improved decoding accuracy.
• Advantages Over Existing Solutions:
  • More accurately predicts the shape of touch point distributions, particularly excelling in 2D rectangular target and screen-originated tasks.
  • Enhanced soft keyboard decoding performance, reducing decoding error rates on smartwatches from 28.89% in the original model to 28.54%.
• Experimental or Evaluation Results:
  1. Model Fitting Experiment:
     • Using Ko et al.'s target acquisition dataset, the Rotational Model achieved the lowest WAIC score, demonstrating the highest predictive capability.
  2. Decoding Experiment:
     • In keyboard decoding tasks on smartwatches and smartphones, the rotational model achieved lower average decoding error rates, especially when user input deviated from the horizontal axis.
  3. Touch Point Distribution Prediction:
     • The model better captured the characteristic alignment of the touch point distribution's major axis with the movement direction compared to the original dual Gaussian model.
• Limitations and Future Directions:
  • Limitations:
    • The rotational model requires knowledge of the finger movement direction θ; if unknown, only the original dual Gaussian model can be used.
    • The current model has only been validated for the index finger; parameter re-estimation is needed for other fingers.
  • Future Directions:
    • Explore optimizing the model to accommodate unknown directions and multi-finger input.
    • Apply the Rotational Dual Gaussian Model to complex graphical interface design and dynamic target task prediction.
    • Investigate the model's potential for predicting touch point distribution in multi-user environments.

Conclusion

The Rotational Dual Gaussian Model significantly enhances the adaptability and accuracy of touch point distribution prediction by incorporating finger movement direction into the covariance matrix construction. The study demonstrates the model's potential applications in target acquisition tasks, soft keyboard decoding, and touchscreen interface design, particularly excelling in 2D target and highly interactive task scenarios.