2. Problem
• To interact with mobile devices on-the-go
• Glasses, wrist band, watch not good
• User have to practice
• Environment limits
• Too heavy processing, power hungry
2
8. Surface detection - touch
• Pressured-based
• When touch or curl,
stiffen and move laterally
• Force sensing resistor
on inner wall of ring
• Consume less power
8
9. Surface detection - motion
• Audio-based
• Friction between fingers and surfaces
• Function of finger motion speed,
and properties of the surface
• Noise
• Use band-pass filter
• They use common office paper
9
11. Energy harvesting
• 10mAh battery capacity
• NFC-enabled phone
• RFID is a type of NFC
• Coil loop around ring band
• Maximize the size
• Position C best
• Over 5% capacity is available
11
13. Future work
• Wireless transmission no detect or recover from
packet loss or error
• Improve classification accuracy
• Improve noise resilience
• Larger variety of surfaces
• Lack of feedback
• Larger amount of training data
13
Other use mobile phone or Nintendo Wii Remote
重classification accuracy
我們重power consumption, perpetual availability
用加速器 紅外線 陀螺儀 camera…
在任何表面上:書桌 椅子把手 衣服…
不用拿下來就可以充電
Tap 4下進入active mode
判斷在寫還是在動
Immediate feedback
Scroll斜的模仿zoom in and out
可以在user指尖裝東西簡單的達到,但是這種方法不能長時間用
按或彎的時候,這個肌肉會用力
但彎不會彎四下引起active mode
High pitched human voice還是有點難去掉noise error
用速度判斷soft or hard landing,落下前有沒有很快的減速
Hard->用touch duration判斷tap swipe
Soft->
Scroll需要real time identification
Stroke可以有時間選字 non real time
Stroke in text entry area, 可能一筆畫很長”L”
cons:
用surface作相對X Y軸,但過程中假設角度不會變
加速器data 沒有yaw方向 因為沒有陀螺儀,假設user motion都會和身體垂直,影響不大
因為想要戒指夠小所以沒辦法裝太大的電池
電磁感應生電
不能讓電量掉到5%以下,不到2.5V
作實驗:每2hr active 1~30min
Last charging and unavailable fail, 4hr以後才會failure 75%以上
找feature看各種threshold怎麼切或取幾個feature per axis
Landing acceleration, tap-swipe duration, stroke classifier
Stroke圓的比直線準,因為斜的和直的容易搞混,user沒有訓練過
More data, more better