Optimization of Tongue Gesture Processing Algorithm for Standalone Multimodal Tongue Drive System
Optimization of Tongue Gesture Processing Algorithm for Standalone Multimodal Tongue Drive System
复制标题
DOI：
10.1109/jsen.2018.2887257
复制
发表时间：
2019-04-01
影响因子：
4.3
通讯作者：
Ghovanloo, Maysam
中科院分区：
文献类型：
--
作者：
Sahadato, Md Nazmus;Sebkhi, Nordine;Ghovanloo, Maysam
多模态舌头驱动系统（mTDS）是一种辅助技术，它使用四肢瘫痪患者的语音，舌头和头部运动来帮助他们控制轮椅，PC或智能手机等设备。为了处理舌头手势，需要求解磁偶极子方程，这在可穿戴嵌入式硬件中执行是计算昂贵的。在本文中，提出了一种基于线性核算法的支持向量机，该算法使用了一个额外的磁力计，在15个参与者中具有96.1%的准确度，2.2%的灵敏度和1.8%的特异性，对mTDS具有计算效率，即使磁体位置发生微小变化，也能产生高性能的分类结果（-5至5 mm）和取向（-10度至10度）的四维机器人模拟舌头运动数据与五种其他算法进行比较。该算法生成更准确的结果，以独立控制设备，尽管语音和舌头运动之间的Midas触摸问题。
Multimodal tongue drive system (mTDS) is an assistive technology, which uses speech, tongue, and head movements of people with tetraplegia to help them control devices such as wheelchairs, PCs, or smartphones. In order to process tongue gestures, it is required to solve the magnetic dipole equation, which is computationally expensive to execute in a wearable embedded hardware. In this paper, a support vector machine with linear kernel-based algorithm is proposed, which uses an additional magnetometer that has 96.1% accuracy with a 2.2% sensitivity and 1.8% specificity among 15 participants, is computationally efficient for mTDS, generates high-performance classification results despite small changes of magnet positions (-5 to 5 mm) and orientations (-10 degrees to 10 degrees) from a four-dimensional robot emulated tongue movement data compared to five other algorithms. This algorithm generates more accurate results to control devices independently despite the Midas-touch problem between speech and tongue movements.