计算机工程与应用
計算機工程與應用
계산궤공정여응용
COMPUTER ENGINEERING AND APPLICATIONS
2015年
18期
199-203
,共5页
视频增强%心率测量%人脸定位%特征点跟踪%实时
視頻增彊%心率測量%人臉定位%特徵點跟蹤%實時
시빈증강%심솔측량%인검정위%특정점근종%실시
video enhancement%heart rate measurement%face detection%feature points tracking%real-time
随着心脏的收缩舒张,人脸部毛细血管中氧合血红蛋白的含量会发生变化。由于氧合血红蛋白对光线有一定的吸收作用,皮肤颜色也会随之发生轻微变化,这种变化裸眼无法识别。利用一种有效的视频增强方法,提出了一种基于视频的心率测量算法。该算法对输入视频进行人脸定位与特征点跟踪,将特征点区域转换为时域信号,对信号进行放大滤波处理,得到心率值。为了提高算法的测量精度和鲁棒性,提出了利用绿色通道、多特征点同时处理。同传统的接触式心率测量设备相比,该算法无需接触人体,有着对人体束缚度小、方便使用的优点。同时,该心率测量算法可实时运行,平均误差仅为3.46次/min,达到实际应用需求。
隨著心髒的收縮舒張,人臉部毛細血管中氧閤血紅蛋白的含量會髮生變化。由于氧閤血紅蛋白對光線有一定的吸收作用,皮膚顏色也會隨之髮生輕微變化,這種變化裸眼無法識彆。利用一種有效的視頻增彊方法,提齣瞭一種基于視頻的心率測量算法。該算法對輸入視頻進行人臉定位與特徵點跟蹤,將特徵點區域轉換為時域信號,對信號進行放大濾波處理,得到心率值。為瞭提高算法的測量精度和魯棒性,提齣瞭利用綠色通道、多特徵點同時處理。同傳統的接觸式心率測量設備相比,該算法無需接觸人體,有著對人體束縳度小、方便使用的優點。同時,該心率測量算法可實時運行,平均誤差僅為3.46次/min,達到實際應用需求。
수착심장적수축서장,인검부모세혈관중양합혈홍단백적함량회발생변화。유우양합혈홍단백대광선유일정적흡수작용,피부안색야회수지발생경미변화,저충변화라안무법식별。이용일충유효적시빈증강방법,제출료일충기우시빈적심솔측량산법。해산법대수입시빈진행인검정위여특정점근종,장특정점구역전환위시역신호,대신호진행방대려파처리,득도심솔치。위료제고산법적측량정도화로봉성,제출료이용록색통도、다특정점동시처리。동전통적접촉식심솔측량설비상비,해산법무수접촉인체,유착대인체속박도소、방편사용적우점。동시,해심솔측량산법가실시운행,평균오차부위3.46차/min,체도실제응용수구。
The content of oxyhemoglobin in human facial capillaries changes with the contraction and relaxation of the heart. With oxyhemoglobin absorption of light, skin color will also vary slightly, which cannot be observed by naked eyes. Using an effective video enhancement algorithm, a video-based heart rate measuring method is proposed. The face in the input video is located and feature points are tracked. Then the regions of the feature points are transformed into time-domain signals. The signals are enlarged and filtered to obtain the heart rate. In order to improve the accuracy and robustness of the detection algorithm, the green channel and multi-feature points are utilized. Compared with the traditional contact heart rate measurement equipment, the algorithm does not require contact with the body and has the advantages of a small degree of human bondage. Meanwhile, the heart rate measurement algorithm can be performed in real time and meets the applicable requirement with an average error of only 3.46 time/minute.