- 學位論文
使用兩台方向上垂直的攝影機進行步態分析的系統
A Gait Analysis System Using Two Orthogonally Arranged Cameras
摘要
人口的老化為近年來世界普遍的現象之一，並且獨居老人比例逐年上升，照護人力成本也跟著上升。為了節省醫療資源費用，並且提升醫生診斷效率，本論文提出一套不必限定行走方向的步態分析監測與評估系統，改善過去方法較不便之處，可運用於家中、醫院、診所等場所，讓使用者了解自我身體健康的狀況。 本論文使用多方向的走路影像，分析人類走路的姿勢，使用步伐寬度變化、頸幹線傾斜角度的變化、行走角度等特徵來評估走路的姿勢。首先透過Otsu的方法取得二值化影像所需的門檻值，再利用二值化影像和二值化背景影像相減後所得到的結果中取出人體輪廓影像進行特徵擷取與分析。利用來自不同行走角度的步伐寬度週期變化，評估走路姿勢是否正常，利用頸幹線傾斜角度的變化，評估身體上半身姿勢是否異常。 本論文成功的擷取多種角度的步伐特徵輪廓影像，並將拍攝角度互相垂直的兩台攝影機所得的特徵值加以整合處理，再與行走方向垂直於攝影機所得到的理想狀態步伐特徵做比較。實驗結果顯示，相關係數可高達89%。證明以兩台視線互相垂直之攝影機進行步態分析，可解決行走方向上的限制，很方便地提供長期的走路姿勢紀錄。
並列摘要
Population aging is one of the common phenomenal in the world. The ratio of elderly living alone increases year by year and the cost of human care follows accordingly as well. To reduce the cost of medical resource and improve the doctor’s diagnosis efficiency, this study proposes a human gait monitoring and evaluation system with no restriction on walking direction. This system improves the inconvenience situation in previous methods and can be applied at home, hospitals, clinics and other places for the self-evaluation of a user’s health condition. This study uses the images obtained from multi-viewing angles and performs human gait analysis through walking features of gait width variation, neck line inclination variation and walking direction angle. First, we obtain the threshold needed for image binarization by using the Otsu’s method. Then we subtract a binary image from the binary background image to get human silhouette image for feature extraction and analysis. We evaluate the gait normality by using the periodicity of gait width variation from various walking directions, and evaluate the normality of upper body by the tilt angle variation of neck line. This study successfully extracts gait feature silhouette images of various walking directions and integrates the features obtained from two orthogorally arranged cameras. The integrated feature is then compared with the ideal gait feature obtained from the camera whose viewing angle is vertical to walking direction, resulting in a high correlation coefficient of 89%. This proves that a vision-based gait analyzer based on two orthogonally arranged cameras can remove the walking direction restriction and provide a long-term gait record conveniently.