改进的Hough与梯度直方图的人眼定位算法  

Improved hough & histogram of oriented gradient for eye location

作　　者：蒋丹丹[1] 冯晓毅[1] 

机构地区：[1]西北工业大学电子信息学院,陕西西安710129

出　　处：《电子设计工程》2014年第21期127-130,共4页Electronic Design Engineering

基　　金：航空基金(20131353015)

摘　　要：针对民用航空器飞行中,实时监控机组座舱行为的疲劳检测系统的快速、精确、低计算量的需求,提出了一种基于Hough、梯度直方图结合的算法:利用Adaboost算法检测人脸之后,运用Hough变换进行人眼初检测,再利用提出的简化梯度直方图特征进行人眼的精确定位。算法将目标区域的简化梯度直方图特征送入SVM分类器,对Hough检测后存在的多个"可能圆"进行筛选,剔除Hough检测中的非人眼圆。实验对比结果表明,改进的人眼定位结合算法能够快速的检测人眼,同时粗定位缩小了Hough圆检测的范围,减小了计算量,缩短了算法运行时间,SVM和改进的梯度信息提高了定位的准确率,并且检测结果稳定、鲁棒性好。Aiming at satisfying the fast, accurate, low computational fatigue detection system which is used to monitor crew cabins behavior in civil aircraft, a new combination eye location algorithm based on Hough, Histogram of oriented gradient and SVM is proposed：after Adaboost algorithm detecting faces, use Hough circle detection algorithm for first-eye-detection, then apply the new simplified Histogram of Oriented Gradient feature to realize accurate eye location. The algorithm sends the simplified histogram of oriented gradient feature of target areas into SVM classifier for selecting right eyes in the “candidate circles” and excluding those not-right eyes. Experimental comparison results show that the improved binding human eyes algorithm quickly detects eyes, the coarse positioning reduces the candidate region of Hough circle detection range, reducing the amount of calculation, shortening the time for the algorithm running, the SVM combined gradient information improves the detection rate, and the test results are stable and good robustness.

关 键 词：民用航空器 疲劳检测系统 人眼定位 改进的结合算法 Hough圆检测 梯度直方图 SVM 

分 类 号：TN911.73[电子电信—通信与信息系统]