平行處理和CPU頻率縮放對於腳底壓力辨識系統之省電研究

Abstract

由於近年來生物辨識技術的興起，讓簽名認證方式不限於以往的帳號密碼，不僅讓生活更為便利且其安全程度也更為可靠。其中，步態辨識在醫療、運動、安全等等領域都有相關研究，我們可以從每個人的腳底獲取許多隱私資訊，根據每個個體不同的運動規律、踩踏重心以及個體大小來進行個體識別。在步態辨識領域中的腳底壓力分析的相關文獻裡，實驗或實作方式大多是以室內插座對電腦進行供電，因此省電方面的研究無人著墨。但當在無插座電源供電的情形下進行應用或實作時，腳底壓力辨識系統就會受到耗電量方面上的限制。因此我們的研究是針對嵌入式系統搭配腳底壓力辨識平台，在沒有室內插座供電的情形下進行省電的研究。我們對部分程式進行平行處理，並從機器學習演算法、CPU頻率模式、和變更核心數的角度對省電比例的影響進行分析，最後針對耗電量進行觀察與解析，並列出了最佳省電和最低耗電量兩種組合。實驗結果顯示，我們所使用的省電方法在四核心訓練階段省電比例可以達到7.02%，辨識階段的省電比例可以達到30.12%。

Due to the rise of biometric technology in recent years, the emerging signature authentication methods, which are not limited to the old-school account password, make life more convenient and more securer. In the relevant literature on plantar pressure analysis, most of the experimental or practical methods use indoor sockets to supply power to the computer, therefore, there is no research on power saving for plantar pressure recognition. However, when applied or implemented in outdoor scenes, the plantar pressure recognition system will be limited by power consumption. Hence, our research aims at the power saving research of the embedded system with the foot pressure recognition platform in the absence of indoor socket power supply. We optimize the programs to run them in parallel, and analyze the impact on the power saving ratio from the perspective of machine learning algorithms, CPU frequency mode, and changing the number of CPU cores. Finally, we observe and analyze power consumption, and list the best combination which can save the highest percentage of power and the combination of the lowest power consumption. The experimental results show that the power saving method we use can reach 7.02% in the four-core training phase, and 30.12% in the four-core identification phase.