眼鏡型デバイスの鼻あてに圧力センサを搭載する耳ぴく入力システム

Wearable devices are aimed to be worn in everyday scenarios, thus the interaction between the human and computer must be quick, accurate, and robust. Button input and touch interfaces are simple and intuitive, however the surface of the wearable devices, such as smartwatches and smart glasses, are small, causing the Fat Finger Problem or need of carrying a second device. In contrast, Hand gestures and vocal input don’t depend on the size of the device. However, the daily movement or environment could cause camera occlusion or catch ambient noise, leading to incorrect input and malfunction. Although specific commands are used to lower the error rates, the input response delays. To address these challenges, this research focuses on using ear wiggling, a vestigial human feature, a movement that does not interfere with daily movements. Ear wiggling provides a fast and voluntary input as a mouse click. Based on this concept, I developed an input method for smart glasses by utilizing pressure sensors on the nose pads. This system is hands-free, eyes-free, resistant to environmental factors, compact, has low power consumption, and is lightweight. The sensing system is more suitable for wearable devices compared to previous EMG methods, thus it doesn’t require skin preparation, precise electrode placement, and isn’t prone to detachment during extended use. This paper is composed of three experiments. Experiment 1 examined the input speed, eight ear wiggling gestures, and NASA-RTLX workload upon five volunteers. The proposed method’s input speed was found to be 36.28ms slower, and the workload was higher than pressing a button with a finger. The gesture waveforms had significant differences in amplitude, number of peaks, and duration depending on the gesture. The workload of the gestures highlighted differences in difficulty, with simple gestures being easy and vice versa. Experiment 2 measured other movements that affect the sensor value. Heart pulse waves, blinking, and walking could be seen, thus a user study was conducted upon seven volunteers to collect data. The data was analyzed using Support Vector Machines and Random Forest classifiers. The per-user classifier achieved a F1 score of up to 98.8 %, while the generic classifier reached 91.4 %, demonstrating the system’s high accuracy and potential for practical application. Experiment 3 trained three volunteers for five days to obtain ear wiggling. None of the volunteers were able to obtain ear wiggling, though we were able to gain points to keep in mind when conducting training. The results will be used to improve the training procedure.