即時抗晃動遠距影像式心跳呼吸量測系統

Abstract

生理資訊 (心跳、呼吸) 為判斷一個人健康狀態的重要指標。傳統量測儀器多為接觸式儀器，使用上會造成許多不方便，例如心律臂帶可能會造成皮膚不適。因此近幾年來開始有遠距影像式生理資訊量測技術之研究，原理為利用連續影像偵測受測者臉部或者是皮膚上的光源變化來量測到心跳和呼吸。然而，目前此種量測方法遇到最大的限制即為受測者量測時靜止不動。因此本論文目標為使受測者能在晃動的情況下也可以使用攝影機量測心跳呼吸，例如身體移動、頭部轉動、跑步、開車、…等等情況。本論文結果顯示在心跳方面能改進受測者晃動的情況，並且在心跳與呼吸方面計算速度皆可以達到30 fps，心跳方面更有專業級比較儀器當作參考基準。 本論文在心跳方面提出了改善人臉偵測之方法，藉由臉部輪廓偵測與低通濾波器、轉動補償等技術增加人臉偵測之穩定度；在運算完心跳時域訊號後，分析其頻譜之特性，設計頻譜峰值選取演算法，克服掉晃動所出現的雜訊正確地找出心跳頻率，並在輸出心跳值前加入穩定機制 在呼吸方面則提出使用SURF演算法偵測胸口位置，使用光流法和濾波器輸出胸口起伏波形，在時域上偵測出起伏頻率以換算出呼吸頻率。

Among all the indicators of health state in humans, heart rate and breath rate are two of the most significant indicators. Conventional contact-type measurement instruments may lead to subjects suffering from the risk of skin irritation and allergic contact reactions. Consequently, recent years have seen increased attention being given to image based remote and contactless physiological signal measuring technique. The main concern of this technique is the potential unreliable result due to the arbitrary motion. As a result, the contribution of this paper is to conquer the motion noise in the following scenario: head motion, body motion, running, driving, etc. Furthermore, the proposed algorithm achieves real time performance up to 30 fps via webcam. So as to prove the effectiveness of proposed algorithm, the ground truth certificated for diagnosis has been utilized. In order to estimate heart rate with motion, several improvement techniques including Face Detection, Peak Chosen, and Protection Algorithm are proposed. The face detection algorithm has been enhanced and stabilized via Face Landmark, position low pass filter and rotation compensation. Peak chosen algorithm in frequency domain is utilized to reduce the noise caused by movement. Lastly, with protection mechanism and Kalman filter, the reliability of heart rate estimation is promising. For estimating the breath rate, SURF feature points detector and matcher are proposed to capture the chest position of subjects. With optical flow calculation and peak chosen algorithm in time domain, breath rate is successfully estimated.