Assessing the hand postures is crucial to implement gesture-based user-computer interfaces for controlling robots and assistive devices. Apart from data gloves and optical tracking, techniques such as surface electromyography and force myography provide a straightforward, non-invasive way to estimate poses and intentions through the forearm muscles assessment. However, most of the myography systems rely on bulky, dedicated hardware with arrays of electrodes or force probes. Therefore, this work introduces the smartphone as an alternative for identifying gestures: with the mobile device attached to the forearm, the embedded inertial measurement unit detects muscular contractions produced during the transitions between postures, yielding signatures in acquired waveforms. After computing the correlation of measured and template patterns, a competitive layer votes the class with greater probability and identifies the gesture. Prior characterization studies evaluated the effect of smartphone placement and forearm orientation in the sensor response, revealing that the IMU signatures are repeatable and robust to positioning deviations. Next, using 10-fold cross-validation, the system discerned four gestures (fist, open hand, wave in, and wave out) performed by six volunteers in ten repetitions, providing 96.6% and 94.1% average accuracies for self-calibration and inter-participant analyses, respectively. The smartphone figures as a ubiquitous and straightforward alternative for assessing gestures, with further applications in human-robot interaction and assistive technologies.

中文翻译：

---

使用智能手机的惯性测量单元识别手势：概念验证研究


评估手势对于实现用于控制机器人和辅助设备的基于手势的用户计算机界面至关重要。除了数据手套和光学跟踪之外，表面肌电图和肌力图等技术还提供了一种直接、非侵入性的方法，通过前臂肌肉评估来估计姿势和意图。然而，大多数肌动描记系统依赖于带有电极阵列或力探针的笨重的专用硬件。因此，这项工作引入了智能手机作为识别手势的替代方案：将移动设备连接到前臂，嵌入式惯性测量单元检测姿势之间转换期间产生的肌肉收缩，从而在获取的波形中产生特征。在计算测量模式和模板模式的相关性后，竞争层以更大的概率对类别进行投票并识别手势。先前的特性研究评估了智能手机放置和前臂方向对传感器响应的影响，表明 IMU 签名可重复且对定位偏差具有鲁棒性。接下来，使用 10 倍交叉验证，系统识别了 6 名志愿者在 10 次重复中执行的四种手势（握拳、张开手、挥手和挥手），为自校准和交互提供了 96.6% 和 94.1% 的平均准确度。 -分别进行参与者分析。智能手机被认为是评估手势的普遍且直接的替代方案，并在人机交互和辅助技术中得到进一步应用。