一种优化的自然光源下眼动视觉测量方法

An optimized visual measurement method for eye movement under natural light source

针对传统的Adaboost算法将非人脸、人眼目标误识别为人脸、人眼的问题,将基于椭圆模型的肤色检测融合进人脸定位算法并采用优化的先验知识法实现人眼定位。采用优化的梯度重心法和Dlib特征点法进行虹膜中心、眼角点定位,完成自然光源下眼动的视觉测量,从而解决传统的虹膜中心、眼角点定位算法存在参数设置依赖于光照条件、图像质量的问题。实验结果表明,该优化算法与传统方法相比,人脸误检率降低了8%,人眼真实检测率提高了21%,虹膜中心定位误差仅相差0.1,视频帧眼角点定位时间降低了55 ms。

In order to solve the problem that traditional Adaboost algorithm misidentifies the targets of face and eyes,the skin color detection based on elliptic model is integrated into the face location algorithm and the optimized prior knowledge method is adopted to realize the eyes location.As the parameter setting of traditional algorithms depends on light condition and image quality,the optimized gradient barycenter method and Dlib feature point method are adopted to locate iris center and canthus point.The experimental results show that,compared with the traditional method,this optimization algorithm reduces the false detection rate of face by 8%,improves the real detection rate of eyes by 21%,makes only 0.1 difference in iris center positioning error,and reduces the positioning time of canthus points of video frame by 55 ms.