基於電阻感測的手勢辨識手環

Abstract

手勢辨識技術是個非常熱門的研究主題，尤其在人機互動 (Human-Computer Interaction, HCI) 的領域中，手勢辨識的技術可以幫助使用者用更符合直覺的方式操控電腦。本研究著重於開發人機互動中的手勢辨識裝置，功能可以是以手勢向智慧助手下指令、或是比較娛樂取向，做為 VR/AR 系統的輸入端。對於我們的目標而言，希望裝置能有幾項特性：穿戴舒適、隨時隨地都能使用、保有隱私不被窺探、製做成本低。 然而目前存在的研究中，還沒有完全符合我們需求的項目。基於電腦視覺和無線訊號偵測的手勢辨識技術，可移動性 (mobility) 較低，而目前有的穿戴式的手勢辨識手環，可穿戴性 (wearability) 還不夠高。舉例來說，基於肌電 (Electromyography, EMG) 感測的裝置，需要黏貼電極片在手臂上、基於彎曲感測器 (flex sensor) 的裝置，需要配戴有感測器的手套在手上，這些裝置的穿戴方法在日常生活中，對使用者都會造成不便。現有的研究中，最接近我們需求的作品，是一條基於電容感測的手環，它可以偵測手腕輪廓的變化，以此辨識手勢。不過這項作品還是有一項缺點，就是手環使用硬質的矽膠製成，在穿戴上還是不夠舒適，而且其手環結構複雜，製做成本較高。 因此，本研究旨在改良此款手環，沿用手腕輪廓辨識的靈感，但探索使用其它種感測器做辨識的可能性。在嘗試了多種材料後，我們發現了導電橡膠這項材料。導電橡膠為混合碳黑 (carbon black) 粉末的橡膠，具有導電性，且電阻值會因為受力而發生改變，因此我們可以測量橡膠的電阻值並回推裝置受力或是變形情形。導電橡膠柔軟、有彈性，在穿戴時不會造成使用者不適，看似非常適合作為HCI裝置使用。但是實際上少有作品使用此材料，理由是此類應力感測器的電阻變化和伸長比例間並非線性關係，據我們所知目前也還沒有人建立出導電橡膠的電阻變化模型。因此在實驗前，我們也不太確定導電橡膠是否真的適合使用。 在本研究中，我們首先對導電橡膠條進行大量的測量，收集導電橡膠條在不同原長和不同伸展長度下的電阻值並建立模型，發現導電橡膠在伸展量很小的時候，會有非常顯著的電阻變化，也因此認定能使用導電橡膠做為手環上的感測器，決定改以電阻感測的方式做辨識。接著，我們以導電橡膠條和丹寧布製做了手環，並利用前述的電阻模型設計校正數據的演算法，再將校正後的數據輸入Random Forest Classifier 辨識使用者做出的手勢。

Hand gesture recognition is a popular research topic, especially in the field of Human-Computer Interaction (HCI). This study focuses on developing a wearable hand gesture recognition device for human-computer interaction, which can be used to give commands to smart assistants through gestures or be utilized as an input interface for VR/AR systems with entertainment purposes. We aim to achieve several characteristics: comfort while wearing, usability anytime and anywhere, privacy preservation, and low cost. However, existing projects do not fully meet our requirements. Technologies based on computer vision and wireless signal detection lack mobility, while current wearable gesture recognition wristbands are not wearable enough. The closest existing work to our requirements is a wristband based on capacitive sensing, capable of detecting changes in wrist contours to recognize gestures. Nonetheless, this work has a drawback as the wristband is made from rigid silicone, leading to discomfort during wear, and it has a complex structure, resulting in higher production costs. Therefore, this study aims to improve this wristband design by retaining the idea of wrist contour recognition but using different types of sensors for recognition. After trying various materials, we discovered a material called conductive rubber. Conductive rubber is soft, elastic, and comfortable to wear, making it seemingly suitable for HCI devices. However, few works have utilized this material due to the nonlinearity between resistance change and elongation proportion. In this study, we first conducted extensive measurements on conductive rubber cord to build a model for the resistance change of it. We found that resistance value exhibits significant changes when subjected to small extensions. Based on this discovery, we decided to use conductive rubber cord as a sensor on the wristband and employ a resistance sensing approach for gesture recognition. We then created the wristband using conductive rubber cords and denim fabric. By employing the previously developed resistance model and a calibration algorithm for the collected data, we fed the calibrated data into a Random Forest Classifier to recognize the gestures made by the user.