基于视频的非接触式生理参数监测关键技术研究

Non-contact monitoring of human physiological parameters (e.g. heartbeat, breathing, etc.) is a hot topic in the field of biomedical engineering. In recent years, video-based monitoring has drawn increasing attention due to its low cost and ease of operation. However, current research is limited to the ideal situations that ambient light is relatively stable and human body is relatively motionless. Under the real-life situations that ambient light is changing and human body is moving, the available methods are unable to make valid estimates of physiological parameters, severely restricting the scope of its applications. This proposal aims to employ advanced video/image processing methods and the recent proposed joint blind source separation techniques to explore the aforementioned problem. Based on particularly designed information fusion strategy, the human facial video signal is used for landmark localization, region segmentation, joint source extraction and statistical analysis. Then, the underlying sources for dynamic lighting, motion artifact, heartbeat, respiration and background noise can be effectively separated and identified. A set of effective solutions will be proposed for robust estimation of physiological parameters under the real-life situation. The launching of this study will significantly promote the practicability of the video-based monitoring technology towards clinical medicine and health care.

人体生理参数（如心跳、呼吸等）的非接触式监测是生物医学工程领域的一个研究热点，近年来，基于人体视频的监测技术由于具有低成本和易操作等特点而得到广泛关注。然而，现有相关研究仅局限于环境光相对稳定和人体相对静止的理想状态，尚不能在环境光变化和人体移动的真实情景下对生理参数作出有效估计，严重制约了其应用范围。本项目拟采用先进的视频图像分析技术和最近兴起的联合盲源分离技术，通过设计有针对性的信息融合策略，对人体视频信号，特别是面部信息，进行特征点定位、区域分割、联合信号提取和统计分析，从而对视频信号中的动态光源、运动伪迹、心跳、呼吸及背景噪声等信息进行有效分离和辨识，提出一套对真实情形鲁棒的非接触式生理参数监测解决方案，进而推动该项技术向临床医学及健康监护的实用化方向发展。

人体生理参数（如心跳、呼吸等）的非接触式监测是生物医学工程领域的一个研究热 点，近年来，基于人体视频的监测技术由于具有低成本和易操作等特点而得到广泛关注。 然而，现有相关研究仅局限于环境光相对稳定和人体相对静止的理想状态，尚不能在环境光变化和人体移动的真实情景下对生理参数作出有效估计，严重制约了其应用范围。本项目采用先进的视频图像分析技术和最近兴起的联合盲源分离技术，通过设计有针对性的信息融合策略，对人体视频信号，特别是面部信息，进行特征点定位、区域分割、联合信号提取和统计分析，从而对视频信号中的动态光源、运动伪迹、心跳、呼吸及背景噪声等信息进行了有效分离和辨识，提出了一套对真实情形鲁棒的非接触式生理参数监测解决方案，有效推动了该项技术向临床医学及健康监护的实用化方向发展。项目共发表高水平学术论文31篇（其中ESI高被引4篇，10篇IEEE期刊文章），其中项目负责人第一/通讯作者论文19篇；申请中国发明专利9项，其中授权2项，专利权转让1项。项目开展期间，项目组成员积极与国内外相关院校和科研院所开展广泛合作与交流，参加国内外相关学术会议。依托此项目，项目执行期间，培养研究生11名。