使用多無線網路訊號源之手勢辨識

Abstract

本論文實作多無線網路訊號源之手勢辨識系統。 無線網路訊號具備可以穿透牆壁的特性，加上現今無線訊號裝置的高普及率等特性，我們希望能夠藉由無線訊號在不同手勢下時會有不同的都卜勒效應變化，藉以實現一個與傳統手勢辨識方法中影像辨識系統中(受相機鏡頭範圍所限制和光源強度影響)、實體感測器(需無時無刻配戴感測器裝置)等不同的手勢辨識系統。 然而，藉由實驗數據發現，手勢辨識正確率會因位置而有顯著的影響。因此，我們提出方法藉由在實際生活情境同一個空間中，可能同時有多無線網路訊號源的特性，像是現今家庭中同時具備手機、平板電腦、筆電$cdots$等等會產生無線訊號的裝置，根據實驗結果我們發現當其中一個訊號源因為手勢位置所產生的都卜勒效應不夠顯著到可以辨別時，可以藉由另一訊號源的能產生較好的都卜勒效應來成功的辨別手勢。 在此論文中，我們提出三種方式利用不同訊號源資料多樣性來改進因手勢位置而造成辨識正確率降低的問題， 我們可以成功地改善辨識正確率達93\%。從此結果可以看出實現一個居家手勢辨識系統是具可行性的。

This thesis presents a multi-source signal source gesture recognition system. We leverage the characteristics of wireless signal, including traverse through the whole home, and high penetration rate to implement a gesture recognition system which is unlike the line of sight and light condition limitation of vision-based system, non device-free physical sensor system. However, according to our experimental measurement studies, the accuracy of single transmitter system depends on the angle of performing gesture(location). Fortunately, in real world, current environments are full of wireless signals transmitted by different devices coming from various angles. For example, the signals sent by one source is very likely to create a stronger Doppler effect than that sent by another source. In thesis, we presents three approaches exploiting this diversity of multiple signal sources to tackle the above location issues, as a result realizing the whole-home gesture recognition in practice.