腳底壓力辨識系統對於穿著不同鞋種的機器學習與特徵組合之研究

Abstract

物聯網應用在近年生活中越來越廣泛，像是智慧型手機、智慧手錶與電腦等，皆讓人類的生活更加便利，為了快速且更安全的身分認證來解鎖相關設備，生物辨識技術扮演了非常重要的角色，此技術相較於傳統文字密碼而言，不易被偽造且安全度較高。在過去的腳底壓力分析的研究中，大多皆以赤腳為主要實驗條件，對於在多鞋種相關的條件下研究較少，其使用成本較高的設備進行研究，因設備成本較高對於腳底壓力辨識技術廣泛的應用較為困難。本論文主要在探討受測者穿著多鞋種的情況下，使用腳底壓力辨識技術搭配機器學習與特徵進行身分辨識，最終分析不同機器學習與多特徵組合之辨識率、訓練時間和鞋種。實驗結果顯示使用隨機森林 (Random Forest, RF)在多鞋種實驗中可以達到最佳辨識率77%，訓練時間為2.83秒是所有機器學習中訓練時間最快；其在單一鞋種實驗中可以達到86%辨識率並發現慣用鞋能有更高辨識率。

The applications of the Internet of Things have become increasingly widespread in our lives over the years, such as smart phones, smart watches and computers, all of which make our lives more convenient. In order to unlock those devices faster and more securely using authentication, biometrics plays a very important role, as it's not easily forged and is more secure than the traditional text passwords. In the past research on the analysis of foot pressure, there were not many studies discussing the subjects who wear different shoes. Those studies are difficult to use in our lives widely because they use the expensive equipment. This thesis mainly explores the impact of the plantar-pressure recognition system's accuracy, features and the model training time under the subjects who wear different shoes. The experiment results show that using the Random Forest method can reach up to 77 % accuracy and the fastest training time at 2.83 seconds. It can even reach up to 86 % ccuracy in the single type of shoes. Besides, we also discover that the subjects who wear their shoes can get higher accuracy.