基于图像处理的船舶值班监控系统的研究
|
摘要
航行安全问题一直是全球关注的热点。在发生的船舶海难事故中,人为因素特别是疲劳被确认为一个主要因素,由于疲劳导致的船舶值班人员操作失误是大多数海难事故的根源。因此,研究基于图像处理的船舶值班视频监控系统,有着重要的现实意义。
首先,本文利用DirectShow中的filter管理器实现了多条链路的管理,进而实现了多个监控画面同屏显示、图像的帧提取、帧存储功能,并将帧格式转换为OpenCV的图像存储格式,方便下一步的处理。
然后对帧图像进行处理,先后用肤色建模和OpenCV实现的动态检测两种方法检测人脸。肤色建模采用皮肤的YCbCr模型将人脸正面图像二值化,初步确定了待检测范围。OpenCV人脸动态检测是基于Haar特征的,通过收集人脸正负样本,运用程序训练出可以识别人脸的Haar特征分类器,并将这些分类器级联起来,产生检测速度快、精度高的人脸分类器,最终在复杂背景下准确地定位、标识出动态人脸。
最后,对监控画面进行扫描,如果在一段时间内画面中没出现值班人员(即无人脸)或者前后帧图像的人脸中心位置在很小范围内变化,则判断无人值班或者值班不力,系统提供声像报警使得值班人员即刻警醒。
本文通过处理监控画面内的图像,检测画面中是否有值班人员在有效工作,并在实验室条件下实现基本功能。该算法精度高,运算速度快,系统实现的成本低,但还有很大的继续研发价值,可以进行系统网络化、智能化、标准化的设计,使得系统功能更加完善,实时性更好。
Navigation safety is a hot issue in all the time, for all over the world. Especially the fatigue, which is considered as one of the human elements, is identified as a major factor to the shipwrecks. The watch-keepers' operational errors caused by fatigue are the root of the most of maritime accidents. Therefore, the research on ship watch-keeping surveillance system based on image processing has important practical significance.
Firstly, the management of multiple links is completed by filter graph manager in DirectShow. Furthermore the multi-channel, frame extraction and frame store are implemented. At the same time, the frame is transformed to the OpenCV format for the next step.
Then the frame is processed with the skin color model and the OpenCV dynamic face detection technique. Hence the image is divided into skin and non-skin part described by binarization with YCbCr skin model, and then the initial range for further face detection is located. The OpenCV dynamic face detection is based on the Haar features. The cascade of boosted classifier based on Haar features could detect dynamic face through collection of positive/negative samples and training. It has high detection accuracy, high calculation speed and the final face location is exact in the complex background.
Finally, the surveillance image is scanned. In a period, if there is no watch-keeper or the distance between two face centers has small change in approximate frames, the surveillance system will give an alarm to warn the watch-keepers immediately.
The surveillance system proposed in this paper could detect whether the watch-keeper is working effectively in use of the processing for surveillance images, and the basic functions of the system are completed under the laboratory conditions. The algorithm has high detection accuracy, high calculation speed and low implement cost. However, it is worth further exploring to achieve the networking, intelligence and standardization and make the system more perfect and better real-time.
首先,本文利用DirectShow中的filter管理器实现了多条链路的管理,进而实现了多个监控画面同屏显示、图像的帧提取、帧存储功能,并将帧格式转换为OpenCV的图像存储格式,方便下一步的处理。
然后对帧图像进行处理,先后用肤色建模和OpenCV实现的动态检测两种方法检测人脸。肤色建模采用皮肤的YCbCr模型将人脸正面图像二值化,初步确定了待检测范围。OpenCV人脸动态检测是基于Haar特征的,通过收集人脸正负样本,运用程序训练出可以识别人脸的Haar特征分类器,并将这些分类器级联起来,产生检测速度快、精度高的人脸分类器,最终在复杂背景下准确地定位、标识出动态人脸。
最后,对监控画面进行扫描,如果在一段时间内画面中没出现值班人员(即无人脸)或者前后帧图像的人脸中心位置在很小范围内变化,则判断无人值班或者值班不力,系统提供声像报警使得值班人员即刻警醒。
本文通过处理监控画面内的图像,检测画面中是否有值班人员在有效工作,并在实验室条件下实现基本功能。该算法精度高,运算速度快,系统实现的成本低,但还有很大的继续研发价值,可以进行系统网络化、智能化、标准化的设计,使得系统功能更加完善,实时性更好。
Navigation safety is a hot issue in all the time, for all over the world. Especially the fatigue, which is considered as one of the human elements, is identified as a major factor to the shipwrecks. The watch-keepers' operational errors caused by fatigue are the root of the most of maritime accidents. Therefore, the research on ship watch-keeping surveillance system based on image processing has important practical significance.
Firstly, the management of multiple links is completed by filter graph manager in DirectShow. Furthermore the multi-channel, frame extraction and frame store are implemented. At the same time, the frame is transformed to the OpenCV format for the next step.
Then the frame is processed with the skin color model and the OpenCV dynamic face detection technique. Hence the image is divided into skin and non-skin part described by binarization with YCbCr skin model, and then the initial range for further face detection is located. The OpenCV dynamic face detection is based on the Haar features. The cascade of boosted classifier based on Haar features could detect dynamic face through collection of positive/negative samples and training. It has high detection accuracy, high calculation speed and the final face location is exact in the complex background.
Finally, the surveillance image is scanned. In a period, if there is no watch-keeper or the distance between two face centers has small change in approximate frames, the surveillance system will give an alarm to warn the watch-keepers immediately.
The surveillance system proposed in this paper could detect whether the watch-keeper is working effectively in use of the processing for surveillance images, and the basic functions of the system are completed under the laboratory conditions. The algorithm has high detection accuracy, high calculation speed and low implement cost. However, it is worth further exploring to achieve the networking, intelligence and standardization and make the system more perfect and better real-time.
引文
[1]李伟弘.船员疲劳对船舶值班的影响与评判[J].船海工程,2007,36(4)：110-113.
[2]徐伯民,秦臻.海上船舶碰撞事故原因探讨—疲劳的剖析[J].中国航海,2009,32(4)：115-120.
[3]朱国锋.海员不良心理因素的表现与成因[J].大连海事大学学报.2001,27(3)：45-48.
[4]陈伟秋,鞠杨,穗交宣.广州撞船事故原因初步查明：目前仍有3名失踪者.http://news.sohu.com/20071125/n253455645.shtml,2007-11-25.
[5]人民网.历史的今天.http://www.people.com.cn/GB/historic/0928/4867.html,1994-9-29.
[6]罗天梁.广东九江大桥615断桥事故.http://news.oeeee.com/channel/1928.html,2007-6-17.
[7]航运信息网.浅谈船舶碰撞事故预防.http://www.csi.com.cn/face/hyzxNews/20070517144458.html,2007-5-17.
[8]张显库,任光,刘军,赵卫军,贾欣乐.综合船舶监控系统设计[J].中国造船,2002,43(2)：71-80.
[9]http://zhidao.baidu.com/question/131280088.
[10]http://www.tribon.cn/html/87/t-62787.html.
[11]http://bbs.81tech.com/simple/?t130041.html.
[12]陆其明.DirectShow开发指南[M].北京：清华大学出版社,2003.
[13]Attila Jozsefkun, Zoltan Vamossy, Johnvon Neumann. Traffic Monitoring with Computer Vision. SAMI 2009-7th International Symposium on Applied Machine Intelligence and Informatics, Proceedings. Herlany, Slovakia,2009:131-134.
[14]范伊红,黄涛,彭海云,吕运朋.基于DirectShow的视频图像处理系统设计与实现[J].计算机与数字工程,2006,10(34)：120-123.
[15]陆其明.DirectShow务实精选[M].北京：科学出版社,2004.
[16]华畯,杨树堂,李建华.基于DirectShow技术视频流捕捉及压缩的实现方案[J].计算机工程，2004，30(12)：143-146.
[17]Na Li,Xiao-Shi Zheng,Yan-Ling Zhao. Video-based Vehicle Detection Scheme in Complex Traffic Scene at Urban Intersection.2009 International Conference on Information Engineering and Computer Science, Wuhan:China,2009.
[18]杨勇,慕德俊,张新家.基于DirectShow框架的实时监测系统的设计与实现[J].信息安全与通信保密,2009,1：93-95.
[19]潘爱民.COM原理与应用[M].北京：清华大学出版社,2000.
[20]王社伟,宋敏.COM组件技术的原理及应用[J].福建电脑,2005(10)：34-35.
[21]赵海春.COM组件的设计与使用[J].邵阳学院学报,2009,6(1)：58-60.
[22]王洋,娄庆英.浅谈COM技术[J].沈阳电力高等专科学校学报,2004，4(2)：73-75.
[23]钟玉琢、向哲、沈洪.流媒体和视频服务器[M].北京：清华大学出版社,2003.
[24]宋玉锋.远程数字视频监控系统的设计与实现[J].计算机工程.2002,20(8)：112-118.
[25]陈胜勇,刘盛.基于OpenCV的计算机视觉技术实现[M].北京：科学出版社,2008.
[26]Sugano Hiroki, Miyamoto Ryusuke. OpenCV implementation optimized for a cell broadband engine processor.2009 IEEE 13th Digital Signal Processing Workshop and 5th IEEE Signal Processing Education Workshop, Marco Island, United States:2009,182-187.
[27]布拉德斯基,克勒.学习OpenCV[M].北京：清华大学出版社,2009.
[28]吴晓阳.基于OpenCV的运动目标检测与跟踪[D].浙江：浙江大学.2008年5月
[29]刘瑞祯,于仕琪.OpenCV教程基础篇[M].北京：北京航空航天大学出版社,2007.
[30]Landre Jerome,Truchetet,Frederic. Optimizing signal and image processing applications using Intel libraries. Eighth International Conference on Quality Control by Artificial Vision, Le Creusot, France:2007.
[31]于仕琪,刘瑞祯.学习OpenCVi(中文版)[M].北京：清华大学出版社,2009.
[32]朱双燕.基于肤色的人脸检测与识别方法的研究[D].武汉：武汉理工大学信息工程学院,2007.
[33]陆宗骐,金登男.Visual C++.NET图像处理编程[M].北京：清华大学出版社,2006.
[34]Ommering R,Linden F,Kramer J.et al.The Koala component model for consumer electronics software.IEEE Computer,2000,33(3):78-85.
[35]段新涛,刘栋.基于肤色和图像似然度的人脸检测[J].中国科技信息,2006(15)：142-143.
[36]张海林,李榕,常鸿森.基于YCbCr模型和形态学的瞳孔分割及人脸检测[J].计算机仿真,2006,23(10)：217-220.
[37]尹星云,时慧坤.数学形态学在灰度图像处理中的理论和应用[J].电脑知识与技术,2006,17：191-192.
[38]尹星云,王峻.基于改进的彩色图像形态学膨胀和腐蚀算子设计[J].计算机工程与应用,2008,44(14)：172-174.
[39]陈烯.基于肤色的人脸检测算法研究[D].天津：天津大学.2005.
[40]王航宇.基于YCbCr高斯肤色模型的人脸检测技术研究[J].图像分析,2008,31(22)：102-105.
[41]沈常宇，许潘园.肤色建模和肤色分割的人脸定位研究[J].光电工程,2007,34(9)：103-108.
[42]郭磊,王秋光.Adaboost人脸检测算法研究及OpenCV实现[J].哈尔滨理工大学学报,2009,14(5)：123-126.
[43]刘礼辉.基于Adaboost的快速人脸检测系统[J].科技风,2009,2(1)：60-61.
[44]刘侠,李苏,李延军.一种改进的Adaboost算法的人脸检测分类器[J].空军工程大学学报(自然科学版),2009,10(2)：76-80.
[45]宋义伟,王秀,赵雪竹,朱雪峰.基于肤色分割和AdaBoost算法的彩色图像的人脸检测[J].自动化与信息工程,2009,1(7)：6-10.
[46]赵黎.基于OpenCV的人脸检测系统设计与实现[J].科技信息,2008,2(18)：351-403.
[47]梁艳,田斐,崔世林.基于VC++.net和OPENCV的身份识别系统开发[J].电气技术,2008,5(6)：33-35.
[48]徐志平,张海朝.基于Haar小波变换和分块DCT的人脸识别[J].微型机与应用,2009,5(21)：25-28.
[49]S. Baker, S.K. Nayar. Pattern rejection.CVPR'96.Proceedings of the Conference on Computer Vision and Pattern Recognition,1996,5(6):544-549.
[50]D. Delgado, L.H. Clemmensen, B. Ersboll, J.M. Carstensen. Precise acquisition and unsupervised segmentation of multi-spectral images.Computer Vision and Image Understanding,2007,106 (2-3):183-193.
[51]P. Viola, M.J. Jones. Rapid object detection using a boosted cascade of simple features.CVPR'01. Proceedings of the Conference on Computer Vision and Pattern Recognition, Los Alamitos,2001,CA, USA:511-518.
[52]杨华.数字视频监控系统的研究与实现[D].大连：大连海事大学.2005年3月.
[53]王耀南,李树涛,毛建旭.计算机图像处理与识别技术[M].北京：高等教育出版社.2000.
[54]Milan Sonka, Vaclav Hlavac, Roger Boyle,艾海舟等译.图像处理、分析与机器视觉(第二版)[M].北京：人民邮电出版社,2003.
[55]蒋杰.新型智能车辆视觉系统及图像处理技术的研究[D].长春：吉林大学,2003年6月.
[56]张泽旭,李金宗,李宁宁.基于光流场分割和Canny边缘提取融合算法的运动目标检测[J].电子学报,2003,3(9)：1299-1302.
[57]J. Wu, J. Rehg, M. Mullin. Learning a rare event detection cascade by direct feature selection.NIPS'04. Proceedings of the Conference on Advances in Neural Information Processing Systems,2004,22(6):855-861.
[58]R. Lienhart, J. Maydt. An Extended Set of Haar-like Features for Rapid Object Detection. IEEE ICIP,2002,2(1):900-903.
[59]Zhang ZhenQiu, Zhu Long. Li Stan Z. Zhang HongJiang, Real-time multi-view face detection. IEEE International Conference on Automatic Face and Gesture Recognition,2002.
[60]Freund Y. An adaptive version of boosting by majority algorithm. Machine Learning,2001,43 (3):293～318.
[2]徐伯民,秦臻.海上船舶碰撞事故原因探讨—疲劳的剖析[J].中国航海,2009,32(4)：115-120.
[3]朱国锋.海员不良心理因素的表现与成因[J].大连海事大学学报.2001,27(3)：45-48.
[4]陈伟秋,鞠杨,穗交宣.广州撞船事故原因初步查明：目前仍有3名失踪者.http://news.sohu.com/20071125/n253455645.shtml,2007-11-25.
[5]人民网.历史的今天.http://www.people.com.cn/GB/historic/0928/4867.html,1994-9-29.
[6]罗天梁.广东九江大桥615断桥事故.http://news.oeeee.com/channel/1928.html,2007-6-17.
[7]航运信息网.浅谈船舶碰撞事故预防.http://www.csi.com.cn/face/hyzxNews/20070517144458.html,2007-5-17.
[8]张显库,任光,刘军,赵卫军,贾欣乐.综合船舶监控系统设计[J].中国造船,2002,43(2)：71-80.
[9]http://zhidao.baidu.com/question/131280088.
[10]http://www.tribon.cn/html/87/t-62787.html.
[11]http://bbs.81tech.com/simple/?t130041.html.
[12]陆其明.DirectShow开发指南[M].北京：清华大学出版社,2003.
[13]Attila Jozsefkun, Zoltan Vamossy, Johnvon Neumann. Traffic Monitoring with Computer Vision. SAMI 2009-7th International Symposium on Applied Machine Intelligence and Informatics, Proceedings. Herlany, Slovakia,2009:131-134.
[14]范伊红,黄涛,彭海云,吕运朋.基于DirectShow的视频图像处理系统设计与实现[J].计算机与数字工程,2006,10(34)：120-123.
[15]陆其明.DirectShow务实精选[M].北京：科学出版社,2004.
[16]华畯,杨树堂,李建华.基于DirectShow技术视频流捕捉及压缩的实现方案[J].计算机工程，2004，30(12)：143-146.
[17]Na Li,Xiao-Shi Zheng,Yan-Ling Zhao. Video-based Vehicle Detection Scheme in Complex Traffic Scene at Urban Intersection.2009 International Conference on Information Engineering and Computer Science, Wuhan:China,2009.
[18]杨勇,慕德俊,张新家.基于DirectShow框架的实时监测系统的设计与实现[J].信息安全与通信保密,2009,1：93-95.
[19]潘爱民.COM原理与应用[M].北京：清华大学出版社,2000.
[20]王社伟,宋敏.COM组件技术的原理及应用[J].福建电脑,2005(10)：34-35.
[21]赵海春.COM组件的设计与使用[J].邵阳学院学报,2009,6(1)：58-60.
[22]王洋,娄庆英.浅谈COM技术[J].沈阳电力高等专科学校学报,2004，4(2)：73-75.
[23]钟玉琢、向哲、沈洪.流媒体和视频服务器[M].北京：清华大学出版社,2003.
[24]宋玉锋.远程数字视频监控系统的设计与实现[J].计算机工程.2002,20(8)：112-118.
[25]陈胜勇,刘盛.基于OpenCV的计算机视觉技术实现[M].北京：科学出版社,2008.
[26]Sugano Hiroki, Miyamoto Ryusuke. OpenCV implementation optimized for a cell broadband engine processor.2009 IEEE 13th Digital Signal Processing Workshop and 5th IEEE Signal Processing Education Workshop, Marco Island, United States:2009,182-187.
[27]布拉德斯基,克勒.学习OpenCV[M].北京：清华大学出版社,2009.
[28]吴晓阳.基于OpenCV的运动目标检测与跟踪[D].浙江：浙江大学.2008年5月
[29]刘瑞祯,于仕琪.OpenCV教程基础篇[M].北京：北京航空航天大学出版社,2007.
[30]Landre Jerome,Truchetet,Frederic. Optimizing signal and image processing applications using Intel libraries. Eighth International Conference on Quality Control by Artificial Vision, Le Creusot, France:2007.
[31]于仕琪,刘瑞祯.学习OpenCVi(中文版)[M].北京：清华大学出版社,2009.
[32]朱双燕.基于肤色的人脸检测与识别方法的研究[D].武汉：武汉理工大学信息工程学院,2007.
[33]陆宗骐,金登男.Visual C++.NET图像处理编程[M].北京：清华大学出版社,2006.
[34]Ommering R,Linden F,Kramer J.et al.The Koala component model for consumer electronics software.IEEE Computer,2000,33(3):78-85.
[35]段新涛,刘栋.基于肤色和图像似然度的人脸检测[J].中国科技信息,2006(15)：142-143.
[36]张海林,李榕,常鸿森.基于YCbCr模型和形态学的瞳孔分割及人脸检测[J].计算机仿真,2006,23(10)：217-220.
[37]尹星云,时慧坤.数学形态学在灰度图像处理中的理论和应用[J].电脑知识与技术,2006,17：191-192.
[38]尹星云,王峻.基于改进的彩色图像形态学膨胀和腐蚀算子设计[J].计算机工程与应用,2008,44(14)：172-174.
[39]陈烯.基于肤色的人脸检测算法研究[D].天津：天津大学.2005.
[40]王航宇.基于YCbCr高斯肤色模型的人脸检测技术研究[J].图像分析,2008,31(22)：102-105.
[41]沈常宇，许潘园.肤色建模和肤色分割的人脸定位研究[J].光电工程,2007,34(9)：103-108.
[42]郭磊,王秋光.Adaboost人脸检测算法研究及OpenCV实现[J].哈尔滨理工大学学报,2009,14(5)：123-126.
[43]刘礼辉.基于Adaboost的快速人脸检测系统[J].科技风,2009,2(1)：60-61.
[44]刘侠,李苏,李延军.一种改进的Adaboost算法的人脸检测分类器[J].空军工程大学学报(自然科学版),2009,10(2)：76-80.
[45]宋义伟,王秀,赵雪竹,朱雪峰.基于肤色分割和AdaBoost算法的彩色图像的人脸检测[J].自动化与信息工程,2009,1(7)：6-10.
[46]赵黎.基于OpenCV的人脸检测系统设计与实现[J].科技信息,2008,2(18)：351-403.
[47]梁艳,田斐,崔世林.基于VC++.net和OPENCV的身份识别系统开发[J].电气技术,2008,5(6)：33-35.
[48]徐志平,张海朝.基于Haar小波变换和分块DCT的人脸识别[J].微型机与应用,2009,5(21)：25-28.
[49]S. Baker, S.K. Nayar. Pattern rejection.CVPR'96.Proceedings of the Conference on Computer Vision and Pattern Recognition,1996,5(6):544-549.
[50]D. Delgado, L.H. Clemmensen, B. Ersboll, J.M. Carstensen. Precise acquisition and unsupervised segmentation of multi-spectral images.Computer Vision and Image Understanding,2007,106 (2-3):183-193.
[51]P. Viola, M.J. Jones. Rapid object detection using a boosted cascade of simple features.CVPR'01. Proceedings of the Conference on Computer Vision and Pattern Recognition, Los Alamitos,2001,CA, USA:511-518.
[52]杨华.数字视频监控系统的研究与实现[D].大连：大连海事大学.2005年3月.
[53]王耀南,李树涛,毛建旭.计算机图像处理与识别技术[M].北京：高等教育出版社.2000.
[54]Milan Sonka, Vaclav Hlavac, Roger Boyle,艾海舟等译.图像处理、分析与机器视觉(第二版)[M].北京：人民邮电出版社,2003.
[55]蒋杰.新型智能车辆视觉系统及图像处理技术的研究[D].长春：吉林大学,2003年6月.
[56]张泽旭,李金宗,李宁宁.基于光流场分割和Canny边缘提取融合算法的运动目标检测[J].电子学报,2003,3(9)：1299-1302.
[57]J. Wu, J. Rehg, M. Mullin. Learning a rare event detection cascade by direct feature selection.NIPS'04. Proceedings of the Conference on Advances in Neural Information Processing Systems,2004,22(6):855-861.
[58]R. Lienhart, J. Maydt. An Extended Set of Haar-like Features for Rapid Object Detection. IEEE ICIP,2002,2(1):900-903.
[59]Zhang ZhenQiu, Zhu Long. Li Stan Z. Zhang HongJiang, Real-time multi-view face detection. IEEE International Conference on Automatic Face and Gesture Recognition,2002.
[60]Freund Y. An adaptive version of boosting by majority algorithm. Machine Learning,2001,43 (3):293～318.