FlexSwipe: Flexible multi-element touch sensing system and temporal shape analysis for complex swipe recognition

• A flexible resistive touch sensor was integrated with ML to capture and classify complex gestures under different conditions. • DTW was used to measure gesture shape similarity by using a warping path. • DTW heatmaps and CWT scalograms served as inputs to ML models. • CNN achieved ∼92 % accuracy for slow swipes and ∼88 % for fast swipes. Advances in complex gesture recognition rely heavily on precise touch sensing patches that accurately interpret user movements. However, reliably distinguishing complex gestures that share similar shapes, durations or pressure variations remain challenging and requires robust technological approaches. This study presents a complete framework that combines a custom-built, flexible, multi-element touch sensing patch with machine learning (ML) to detect and classify complex gestures such as “circular_swipe”, “zigzag_swipe”, “figure_eight”, and “spiral_swipe”. High resolution time-series signals were captured from multiple sensing elements by executing slow and fast swipe gestures. Dynamic Time Warping (DTW) was used for analyzing shape similarity and heatmap generation, while Continues Wavelet Transform (CWT) was applied for time–frequency feature extraction. DTW-heatmaps and CWT-scalograms were used as input features for ML models to learn gesture-specific energy patterns. While traditional ML algorithms performed reasonably well, Convolutional Neural Network (CNN) demonstrated a more robust capability for dynamic gesture classification. Confusion matrices revealed that the system achieved an accuracy of 92 % and 88 % for slow and fast swipe gestures, respectively. Our work demonstrates the potential of combining signal shape analysis with deep learning for advancing touch-based assistive technologies that enable low-effort, intuitive control for individuals with limited mobility.