GoosebumpsEdge: Wearable On-Device Skin Imaging for Piloerection Detection

Abstract

Goosebumps, or piloerection, provide a visible physiological cue that can reflect affective and thermoregulatory responses. We present GoosebumpsEdge, a reproducible ecosystem for wearable skin-surface imaging and lightweight on-device goosebumps detection. Our system consists of (1) a stable imaging hardware jig assembled from four off-the-shelf parts and 3D-printable components, (2) an AtomS3R-CAM firmware for robust image capture with optional Inertial Measurement Unit (IMU) logging, (3) a real-time labeling Collector, and (4) an experiment-facing Hub that publishes goosebumps events via Open Sound Control (OSC). Pilot data from three participants (10 sessions total) indicates that goosebump-related signals are reliably reflected in simple texture features, including FFT-based and Laplacian-based measures. Our approach reliably distinguished target patterns (goosebump events) from background data. Using a Local Binary Pattern (LBP) histogram with L2-regularized logistic regression, we achieved an ROC-AUC of 0.84 and an average precision of 0.65 on a held-out session, with a median ROC-AUC of 0.74 across labeled sessions. The on-device pipeline runs in real time at approximately 5 FPS and converts per-frame probabilities into discrete events using EMA smoothing, hysteresis, and optional z-score-based session adaptation. Together, results demonstrate the feasibility of real-time, on-device goosebumps detection using wearable imaging. By releasing all hardware designs and software components as open source, we aim to support further exploration of piloerection as an embodied physiological signal in augmented human systems.