基于驾驶员视线跟踪的安全驾驶监测

发布时间：2018-01-10 05:15

本文关键词：基于驾驶员视线跟踪的安全驾驶监测　出处：《南昌大学》2016年硕士论文　论文类型：学位论文

【摘要】：随着汽车的普及,对驾驶员的驾驶视线进行跟踪对提高行车的安全性有着重要的意义。视线跟踪也是现阶段备受关注的研究领域,其在人机交互、军事、市场营销、工业控制等领域有着广大的运用前景。本文针对实时性、稳定性、准确性提出了一种非入侵式的视线跟踪框架。使用普通的单摄像头对驾驶员进行驾驶图像采集,利用驾驶员的内眼角点作为参考点,虹膜中心作为动点,通过二者的位置关系建立视线跟踪模型。本系统先对视频获取的图像进行滤波与对比增强处理,在滤波处理中提出中值滤波与双边滤波相结合的方法,并使用一种降维的方式进行权值计算。为了减少搜索空间和提高特征提取的准确性,本系统采用从粗到细的搜索策略。先从预处理后的图像中确定人脸区域,在从人脸区域确定眼睛区域,最后获取虹膜中心和角点特征。针对人脸识别模块,本文提出了一种提出基于先验知识与统计方法相结合的思路,先使用肤色模型确定候选人脸区域,再使用Adaboost算法对候选区域进行检测。该方法不仅缩短了检测时间而且提高了检测的准确率。在获取人脸区域后使用“三庭五眼”加基于灰度值的投影方法确定人眼位置。对于虹膜中心检测本文提出了一种基于中垂线的Hough变换的改进方法,把三维问题空间转换成了二维空间。在角点检测模块,提出使用Harris算子二次递归的算法,进行更精准、更细致的眼角检测。最后,本文基于openCV二次开发,对本文提出的框架与相关改进算法进行测试。结果表明,本系统的框架与相关算法是比较有效的。
[Abstract]:With the popularization of automobile, tracking the driver's line of sight plays an important role in improving the safety of driving. Line of sight tracking is also one of the most concerned research fields at this stage, which is in man-machine interaction and military affairs. Marketing, industrial control and other fields have a broad application prospects. This paper aimed at real-time, stability. In this paper, a non-invasive visual tracking framework is proposed. The common single camera is used to capture the driver's driving image, the driver's inner eye corner is used as the reference point, and the iris center is used as the moving point. The system firstly filters and contrasts the image acquired by video, and puts forward the method of combining median filter and bilateral filter in the filtering process. In order to reduce the search space and improve the accuracy of feature extraction, the system adopts a coarse to fine search strategy. Firstly, the face region is determined from the pre-processed image. Finally, the iris center and corner features are obtained from the face region. For the face recognition module, a combination of prior knowledge and statistical method is proposed in this paper. First use the skin color model to determine the candidate's face area. Then Adaboost algorithm is used to detect candidate regions. This method not only shortens the detection time but also improves the accuracy of detection. Add the projection method based on gray value to determine the position of human eye. For iris center detection, we propose an improved method of Hough transform based on midline perpendicular. The three-dimensional problem space is converted into two-dimensional space. In the corner detection module, a quadratic recursive algorithm using Harris operator is proposed for more accurate and meticulous eye angle detection. Finally. Based on the secondary development of openCV, this paper tests the proposed framework and related improved algorithms, and the results show that the framework and related algorithms of the system are more effective.
【学位授予单位】：南昌大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：U463.6;TP391.41

【相似文献】