基于图像技术的人体坐姿监测分析-human sitting posture monitoring analysis based on image technology.docxVIP

摘要随着社会的发展，电脑已经跟我们的生活密切相关，我们需要用电脑来工作、 学习、娱乐，而由于不正确的坐姿导致电脑使用者的健康损害问题也日益突出， 其中近视、脊椎疾病、肌肉劳损等已经严重影响了人们的正常生活。针对以上问 题，本文对基于数字图像处理技术的坐姿监测问题进行了研究。论文的系统设计目标是：1. 评估当前人体坐姿与良好坐姿之间的差异；2. 判 别出是哪一种不良坐姿，并提示用户加以纠正。本文根据坐姿原理，对常见的六 种坐姿进行了判别，分别是：头部倾斜、身体右倾、身体左倾、身体前倾、身体 后仰、脊椎前曲。从当前计算机结构和使用情况来看，一般情况，摄像头在电脑 显示器的正上方，且人也坐在电脑的正前方，所以本文取正面人体图像进行研究。 在监测前，用户通过系统获取一张良好的坐姿图像作为后续监测的标准，后续监 测时通过比较监测图像与标准图像之间的差异程度来进行判别，并指出用户错误 坐姿类型，提示用户有针对性加以纠正。系统处理过程为：系统每隔一段时间从 视频流中获取二帧图像，经预处理和定位人脸、眼睛、嘴后，根据眼睛的位置判 别坐姿是否有头部倾斜、身体右倾或身体左倾的情况；如果没有三者之一，则定 位肩膀位置，再结合嘴的位置和人脸面积等信息进行身体前倾、身体后仰、脊椎 前曲的判别。如果连续 12 次（每次隔 10s）监测到人体坐姿是同一种错误，则提 示用户加以纠正。本文涉及的关键技术有：基于肤色的人脸区域粗分割，基于二值化法的眼睛、 嘴的检测，基于对称性的肩膀检测。本文通过实验对所设计的坐姿监测系统进行 了测试，证明了其有效性。关键词：坐姿监测 图像处理 人脸检测 肩膀检测 二值化AbstractWith the rapidly development of the modern society, computer has closely related to our daily life, which accompanies us in our work, study and entertainment. This leads a serious problem that the incorrect sitting posture does harm to the health of computer users, e.g. myopia, spondylopathy and muscle strain has seriously influenced the computer user’s life. Thus, the thesis does a research on the monitoring of sitting posture based on the digital image processing technology.The thesis aims at two goals: the first, assessing the difference between the current sitting posture and the good sitting posture, the second, caution the computer users to correct his bad posture. Six usual bad postures have been brought into the monitoring system based on the sitting posture theory, and they are: head inclination, body leaning right, body leaning left, body leaning forward, body leaning backward and spinal bending forward. Learning from the structure about computer with a camera located right above the screen and users’ working habits, generally, users are sitting right in the front of computer. Therefore, this thesis practices the research on frontal image. Firstly, a good sitting posture should be taken as the reference giving the comparable data guiding subsequent monitoring.