眨眼检测与眼睛跟踪

提出一种在视频人脸图像序列中，进行眼睛检测，跟踪和睁、闭状态判别的方法。通过眨眼检测，对眼睛进行定位；使用针对性强的内眼角提取算子，确定内眼角精确位置；利用内眼角特征来动态跟踪眼睛；睁眼模板在线生成和更新，通过当前眼睛区域和睁眼模板的相关分析来判别睁、闭眼状态。实验结果表明，算法在50场/s的处理速度下，内眼角点定位准确率达到98%以上，眨眼检测正确率为97.5%。

Blink detection and eye tracking

A method was proposed to detect and track eyes and to recognize the opening and shutting of eyes in a realtime video stream. Blink detection was used for eyes positioning. An inner eye comer filter was used to locate the positions of inner eye comer accurately. The eye was tracked in real time by detecting the inner eye comer. Open eye template could be created online and dynamically updated. The opening or shutting of eyes could be recognized using correlation with an online open eye template. Experimental results indicate that the proposed algorithm can locate the positions of inner eye corner with an accuracy rate of more than 98%. An accuracy rate of 97.5% in blink detection is obtained using the proposed method at a processing rate of 50 fields per second.