腳底壓力辨識系統結合機器學習之分析與研究

Abstract

由於近年來生物辨識技術的興起，讓認證方式不再同於以往的帳號密碼，不僅使生活更為便利且其安全程度也更為可靠，不過在廣大的生物辨識市場之中，系統成本與辨識度考量下要如何達成平衡一直都是辨識系統難以普及化的重點議題之一，在過去研究發現，系統在特徵提取的結果與系統著重於機器學習效果的比例較少，在訓練時所耗費的成本也較無研究。本論文主要在於研究探討分析腳底壓力資訊取出特徵，並與機器學習搭配組合，創造出快速取得腳底壓力資訊且快速訓練且擁有高準確率的系統模組，接著並進一步根據系統辨識率與感測器感測狀況來調整數量達到節省成本的目的。實驗結果顯示我們所開發的系統不僅在辨識結果上有不錯的成績，在訓練處理時時間與辨識時間上也能達到良好的效果，成本上也比先前的便宜，並獲得對此系統普及化與實作上有助的資訊。In recent years, the biometric technology has become more and more popular, and this makes authentication method is no longer limited to the way which only using account and password. It makes our life become not only convenient but secure. However, in the mass biometric market, the balance of system cost and identification accuracy are always been one of the key issues that the identification system is difficult to popularize. Previous study shows, the rate of using feature extraction and machine learning is not taking the high proportion than now. And there are few studies in research of the cost of data training. Because of this, our study is mainly for research and analyze the characteristics of foot pressure information, combining with machine learning to create a system module that can quickly train data and perform a high accuracy. And then based on the system identification accuracy and the sensor sensing condition, we can adjust the number of sensors to achieves the goal of cost saving. The result shows that the system not only performs good results in the identification accuracy but also saves the training and identification time. The cost is also more cheaper than before, and it will make the biometric technology become more popularization.