Neural networks for efficient touch detection on capacitive panels

Abstract

The advancement of capacitive-based touch panel
technologies has opened new opportunities for their incorporation
into embedded devices. However, this progress also underscores
the need for improved software algorithms to achieve high
precision in calculating touch coordinates. Traditional position
calculation methods often exhibit diminished accuracy when
applied to smaller panels, and modifying and tuning these methods
can be time-consuming and labor-intensive. To address these
limitations, this study investigates the performance of two neural
network architectures, specifically a two-layer fully connected
neural network and a radial basis function network, in enhancing
the accuracy of touch coordinate calculation. A key advantage of
these models is their ability to learn efficiently from limited
datasets while minimizing the risk of overfitting. The high touch
position accuracy achieved by the proposed neural network
solutions makes them suitable for deployment in devices with
limited computing resources, such as microcontrollers.
Furthermore, the simplicity of the proposed models enables their
implementation in embedded systems with low power
consumption, offering a practical and scalable solution for a wide
range of applications. Overall, the integration of these neural
network models in touch coordinate processing provides notable
benefits in terms of accuracy, efficiency, and adaptability.