Estimating the Finger Orientation on Capacitive Touchscreens Using Convolutional Neural Networks
summary
Provides a ground-truth dataset (≈500k labeled touches from 33 participants tracked with motion capture) and shows raw capacitive matrices fed to models in TensorFlow outperform handcrafted features. A three-layer CNN with L2 regularization (32/72/160 filters, 2,000-unit FC, two outputs) reduces error by 8.9% for pitch and 45.7% for yaw versus the prior Gaussian-process baseline. Experiments compare k-NN, random forest, DNN and CNN; model is trained off-phone and source plus weights are on GitHub for porting.
AI-written summary · about
chapters (4)
related talks
16mInvestigating the feasibility of finger identification on capacitive touchscreens using deep ...
18mInfiniTouch: Finger-Aware Interaction on Fully Touch Sensitive Smartphones
08mDesigning a Personalised Sensor Glove Using Deep-Learning
14mDeepFisheye: Near-Surface Multi-Finger Tracking Technology Using Fisheye Camera
19mGroup Touch: Distinguishing Tabletop Users in Group Settings via Statistical Modeling of Touch Pairs
32mIntegrating CNN Models with Flutter Using TensorFlow Lite
14mBack-Hand-Pose: 3D Hand Pose Estimation for a Wrist-worn Camera via Dorsum Deformation Network
21mRecognizing Human features using Deep Networks