彩色空间下运动目标检测与跟踪问题研究
|
摘要
运动目标检测与跟踪是图像处理技术及计算机视觉研究的核心内容之一,广泛应用于视频监控、导航制导、医疗诊断等领域。对目标检测与跟踪方面的研究有着重要的理论意义和实用价值。
基于图像灰度值的目标检测算法设计简单,应用广泛,但灰度特征下图像的色彩成分被掩盖,因此色彩空间下目标检测日益成为研究热点。本文提出了一种基于RGB分量自适应调节的目标检测方法,利用像素点红绿蓝三分量值作为特征信息组,根据背景图像的色彩组成,自适应调节三组分量所占的比重；背景差分时,分别计算三组差值,将三组差分值的绝对值之和作为像素点的差值；利用自适应阈值对差值图像进行考察,进而实现对目标的二值化分离；为了增强复杂背景下的检测效果,对均值背景法进行改进,利用去除目标区域的均值背景法建立背景模型。
Mean Shift跟踪算法原理简单,实时性好,但当目标出现被遮挡的情况时容易失去跟踪目标。为此,引入卡尔曼滤波器,通过考察上一时刻对当前时刻的估计值与当前时刻的观察值,对状态变量进行更新、调整,并进一步求出此刻对下一刻的估计值。然后采用Mean Shift对预测的区域进行迭代计算。则可以对完全遮挡下的目标进行连续稳定的跟踪,显著提高了跟踪效果。
As one of the core researches in the image processing and computer vision technology researches, moving target detection and tracking has been widely applied in many fields, such as video surveillance, navigation guidance, medical diagnostics, and so on.
The target detection algorithm based on gray value can be designed easily and used widely, but the color components of the image are covered under the gray feature. Therefore, the target detection under the color space is becoming a research hot spot.
This thesis presents an adaptive target detection method that is based on the RGB component. This method uses the RGB pixel value as the third feature information group, and it adjusts the proportion of the three components adaptively according to the background color of the image composition. When the backgrounds are different, this method calculates the difference between the three groups, and regards the sum of the absolute difference value of the three groups as the difference between pixels.This method also uses the adaptive threshold for difference image to inspect in order to achieve the target binary separation. To enhance the detection effect under the complex background, this method improves the mean background method, and it also builds the background model by the use of average of the backboard removal method of the target area.
Mean Shift tracking algorithm is simple in principle and real-time, but when the target is blocked, it is easily to lose the track of the target. Therefore, the state variable is updated and modified and the estimated value of the current time to the next moment is calculated by introducing the Kalman filter to examine the estimated value of the previous time to the current time and the observed value of the current time; then the adoption of Mean Shift to carry on iterative calculation towards the predicted area can track the target under complete block continuously and stably, which promotes significantly the tracking effects.
基于图像灰度值的目标检测算法设计简单,应用广泛,但灰度特征下图像的色彩成分被掩盖,因此色彩空间下目标检测日益成为研究热点。本文提出了一种基于RGB分量自适应调节的目标检测方法,利用像素点红绿蓝三分量值作为特征信息组,根据背景图像的色彩组成,自适应调节三组分量所占的比重；背景差分时,分别计算三组差值,将三组差分值的绝对值之和作为像素点的差值；利用自适应阈值对差值图像进行考察,进而实现对目标的二值化分离；为了增强复杂背景下的检测效果,对均值背景法进行改进,利用去除目标区域的均值背景法建立背景模型。
Mean Shift跟踪算法原理简单,实时性好,但当目标出现被遮挡的情况时容易失去跟踪目标。为此,引入卡尔曼滤波器,通过考察上一时刻对当前时刻的估计值与当前时刻的观察值,对状态变量进行更新、调整,并进一步求出此刻对下一刻的估计值。然后采用Mean Shift对预测的区域进行迭代计算。则可以对完全遮挡下的目标进行连续稳定的跟踪,显著提高了跟踪效果。
As one of the core researches in the image processing and computer vision technology researches, moving target detection and tracking has been widely applied in many fields, such as video surveillance, navigation guidance, medical diagnostics, and so on.
The target detection algorithm based on gray value can be designed easily and used widely, but the color components of the image are covered under the gray feature. Therefore, the target detection under the color space is becoming a research hot spot.
This thesis presents an adaptive target detection method that is based on the RGB component. This method uses the RGB pixel value as the third feature information group, and it adjusts the proportion of the three components adaptively according to the background color of the image composition. When the backgrounds are different, this method calculates the difference between the three groups, and regards the sum of the absolute difference value of the three groups as the difference between pixels.This method also uses the adaptive threshold for difference image to inspect in order to achieve the target binary separation. To enhance the detection effect under the complex background, this method improves the mean background method, and it also builds the background model by the use of average of the backboard removal method of the target area.
Mean Shift tracking algorithm is simple in principle and real-time, but when the target is blocked, it is easily to lose the track of the target. Therefore, the state variable is updated and modified and the estimated value of the current time to the next moment is calculated by introducing the Kalman filter to examine the estimated value of the previous time to the current time and the observed value of the current time; then the adoption of Mean Shift to carry on iterative calculation towards the predicted area can track the target under complete block continuously and stably, which promotes significantly the tracking effects.
引文
[1]赵赛君.视频图像序列中的运动目标检测与跟踪研究[D].哈尔滨工业大学,2011.
[2]潘武.基于机器视觉的工件的识别和定位[D].北京化工大学,2012.
[3]Collins R, Lipton A J, Kanade T, et al. A system for video surveillance and monitoring: VSAM final report. Technical Report:CMU-R1-TR-00-12, Camegie Melon University, Pittsburgh, Peen, Ameriea,2000.
[4]Tai J, Tseng S, Lin C, et al. Realtime image tracking for automatic traffic monitoring and enforcement application. Image and Vision Computing,2004,22(6):485-501.
[5]李仁杰,余松煜,熊红凯Spatio-temporal Compensation Based Object Detection for Video Surveillance Systems[J]. Journal of Donghua University(English Edition),2008,(02): 123-129.
[6]韩冰,林明星,丁凤华.机器视觉技术及其应用分析[J].农业装备与车辆工程,2008,(10)：24-27.
[7]李雄,刘允才.视觉机制研究对机器视觉的启发示例[J].中国图象图形学报,2013,02：152-156.
[8]COOPER Dave E. Intelligent transportation systems for smart cities:a progress review[J]. Science China(Information Sciences),2012,12:2908-2914.
[9]Yunji ZHAO,Hailong PEI. Object tracking based on particle filter with discriminative features[J]. Journal of Control Theory and Applications,2013,01:42-53.
[10]Matthews I,I shikawa T, Baker S. The template update problem [J]. IEEE Transaction On Pattern Analysis and Machine Intelligence,2004,26(6):810-815.
[11]陈炳权,刘宏立,孟凡斌.数字图像处理技术的现状及其发展方向[J].吉首大学学报(自然科学版),2009,(01)：63-70.
[12]汪冲,席志红,肖春丽.基于背景差分的运动目标检测方法[J].应用科技,2009,(10)：16-18+30.
[13]王晓凯,李锋.改进的自适应中值滤波[J].计算机工程与应用,2010,(03)：175-176+218.
[14]赵建珍,董增寿.改进的自适应均值滤波算法及应用[J].太原科技大学学报,2010,(01)：34-37.
[15]张元元,吴晓娟,阮秋琦Probabilistic Model-Based Silhouette Refinement for Gait Recognition[J]. Journal of Shanghai Jiaotong University(Science),2010,(01):24-30.
[16]孙少林,马志强,汤伟.灰度图像二值化算法研究[J].价值工程,2010,(05)：142-143.
[17]张毅刚,曹阳,项学智.静态背景差分运动目标检测研究[J].电子测量与仪器学报,2010,(05)：494-499.
[18]A Survey to View Update Problem[J]. Journal of Electronic Science and Technology,2010, (04)318-327.
[19]Wiener model identification and nonlinear model predictive control of a pH neutralization process based on Laguerre filters and least squares support vector machines[J]. Journal of Zhejiang University-Science C(Computers & Electronics),2011,(01):25-35.
[20]Approximation of the Nearest Common Fixed Point of Asymptotically Nonexpansive Mappings in Banach Spaces[J]. Communications in Mathematical Research,2011, (04): 369-377.
[21]Stephen Alstrup,Cyril Gavoille.Haim Kaplan,Theis Rauhe. Nearest Common Ancestors:A Survey and a New Algorithm for a Distributed Environment [J] Theory of Computing Systems,2004,(01):68-71.
[22]李大杰.彩色图像信息的提取与分析在道路识别中的应用[D].重庆大学,2008.
[23]曾俊.图像边缘检测技术及其应用研究[D].华中科技大学,2011.
[24]沈春裕.自适应阈值的运动目标检测技术研究[D].浙江工业大学,2012.
[25]蒋东升.基于数学形态学的边缘检测算法研究[D].电子科技大学,2012.
[26]Joshi Shantanu H,Srivastava Anuj. Intrinsic Bayesian Active Contours for Extraction of Object Boundaries in Images.[J]. International journal of computer vision,2009,81(3): 455-459.
[27]曾脉,左志宏.一种准确而快速的运动目标检测算法[J].成都信息工程学院学报,2008,23(4)：370-373.
[28]王建华.基于嵌入式系统的机器视觉及其在机器人控制中的应用[D].重庆：重庆大学硕士学位论文,2003.12.
[29]曲巨宝,林宏丛,梁洪涛,刘胜.运动图像快速跟踪技术研究[J].重庆师范大学学报,2011,28(1):44-48.
[30]张中方,李金屏,拜佩.基于多颜色空间融合的移动目标检测算法[J].济南大学学报(自然科学版),2011,(02)：191-195.
[31]赵欣,陈峰,吴立知.一种改进的meanshift运动目标跟踪算法[J].通信技术,2011,(11)：76-79.
[32]庞国瑞,葛广英,刘洪朋,肖新帅.基于改进的自适应混合高斯背景模型的车辆检测方法[J].聊城大学学报(自然科学版),2012,(01)：37-41.
[33]杨红霞,杭亦文,刘旭.基于Meanshift与Kalman的视频目标跟踪算法[J].武汉理工大学学报(信息与管理工程版),2012,(02)：147-150.
[34]王田,刘伟宁,韩广良.杜超,刘恋.基于改进MeanShift的目标跟踪算法[J].液晶与显示,2012,(03)：396-400.
[35]姚放吾,许辰铭.基于目标质心的Meanshift跟踪算法[J].计算机技术与发展,2012,(06)：104-106+110.
[36]卢官明,郎苏娟.基于YC bC r颜色空间的背景建模及运动目标检测[J].南京邮电大学学报(自然科学版),2009,06：17-22.
[37]徐晓军,纪玉波.基于HSV颜色空间静止背景中运动目标的检测算法[J].辽宁石油化工大学学报,2010,01：71-75.
[38]丁莹,钱锋,范静涛,姜会林.基于不同颜色空间的运动目标检测算法分析[J].长春理工 大学学报(自然科学版)，2012,04：1-4.
[39]徐以美,郭宝龙,陈龙.基于RGB颜色空间的减背景运动目标检测[J].计算机仿真,2008,03:214-217.
[40]沈盼盼,樊丰,伍瑞卿.基于RGB三通道分离的运动目标检测方法[J].电视技术,2012,36(3)：137-140.
[41]CHEN Jie,MA Tao,CHEN WenJie,PENG ZhiHong. Unsupervised robust recursive least-squares algorithm for impulsive noise filtering[J]. Science China (Information Sciences),2013,(04)203-212.
[42]元辉,李志斌,刘微.基于色度分割与图割算法的视差估计算法[J].华南理工大学学报(自然科学版),2013,(02)：12-18.
[43]雷禹.基于自适应背景替换的运动对象检测算法研究[D].广西大学,2012.
[44]艾凯文.基于自适应混合高斯模型的运动目标检测[D]广西大学,2012.
[45]秦臻,曹剑中. 基于方向直方图的Mean shift目标跟踪新算法[J].电子设计工程.2011,(12)：21-25.
[46]王宇翔,韩振铎,王宏敏,臧传吉.一种基于自适应背景预测的红外弱小目标检测方法[J].激光与红外,2012,(10)：1181-1186.
[47]胥欣,江登表,李勃,陈启美.混合高斯模型运动检测算法优化[J].计算机应用研究,2013.
[48]常向魁.视频运动目标跟踪算法研究[D].河南大学,2007.
[49]田莘.基于MeanShift算法的目标跟踪问题研究[D].西安科技大学,2010.
[50]亢朋朋,朱锡芳,南东亮,金运策.一种改进Mean shift算法的视频监控系统[J].微电子学与计算机,2013,02：153-156.
[2]潘武.基于机器视觉的工件的识别和定位[D].北京化工大学,2012.
[3]Collins R, Lipton A J, Kanade T, et al. A system for video surveillance and monitoring: VSAM final report. Technical Report:CMU-R1-TR-00-12, Camegie Melon University, Pittsburgh, Peen, Ameriea,2000.
[4]Tai J, Tseng S, Lin C, et al. Realtime image tracking for automatic traffic monitoring and enforcement application. Image and Vision Computing,2004,22(6):485-501.
[5]李仁杰,余松煜,熊红凯Spatio-temporal Compensation Based Object Detection for Video Surveillance Systems[J]. Journal of Donghua University(English Edition),2008,(02): 123-129.
[6]韩冰,林明星,丁凤华.机器视觉技术及其应用分析[J].农业装备与车辆工程,2008,(10)：24-27.
[7]李雄,刘允才.视觉机制研究对机器视觉的启发示例[J].中国图象图形学报,2013,02：152-156.
[8]COOPER Dave E. Intelligent transportation systems for smart cities:a progress review[J]. Science China(Information Sciences),2012,12:2908-2914.
[9]Yunji ZHAO,Hailong PEI. Object tracking based on particle filter with discriminative features[J]. Journal of Control Theory and Applications,2013,01:42-53.
[10]Matthews I,I shikawa T, Baker S. The template update problem [J]. IEEE Transaction On Pattern Analysis and Machine Intelligence,2004,26(6):810-815.
[11]陈炳权,刘宏立,孟凡斌.数字图像处理技术的现状及其发展方向[J].吉首大学学报(自然科学版),2009,(01)：63-70.
[12]汪冲,席志红,肖春丽.基于背景差分的运动目标检测方法[J].应用科技,2009,(10)：16-18+30.
[13]王晓凯,李锋.改进的自适应中值滤波[J].计算机工程与应用,2010,(03)：175-176+218.
[14]赵建珍,董增寿.改进的自适应均值滤波算法及应用[J].太原科技大学学报,2010,(01)：34-37.
[15]张元元,吴晓娟,阮秋琦Probabilistic Model-Based Silhouette Refinement for Gait Recognition[J]. Journal of Shanghai Jiaotong University(Science),2010,(01):24-30.
[16]孙少林,马志强,汤伟.灰度图像二值化算法研究[J].价值工程,2010,(05)：142-143.
[17]张毅刚,曹阳,项学智.静态背景差分运动目标检测研究[J].电子测量与仪器学报,2010,(05)：494-499.
[18]A Survey to View Update Problem[J]. Journal of Electronic Science and Technology,2010, (04)318-327.
[19]Wiener model identification and nonlinear model predictive control of a pH neutralization process based on Laguerre filters and least squares support vector machines[J]. Journal of Zhejiang University-Science C(Computers & Electronics),2011,(01):25-35.
[20]Approximation of the Nearest Common Fixed Point of Asymptotically Nonexpansive Mappings in Banach Spaces[J]. Communications in Mathematical Research,2011, (04): 369-377.
[21]Stephen Alstrup,Cyril Gavoille.Haim Kaplan,Theis Rauhe. Nearest Common Ancestors:A Survey and a New Algorithm for a Distributed Environment [J] Theory of Computing Systems,2004,(01):68-71.
[22]李大杰.彩色图像信息的提取与分析在道路识别中的应用[D].重庆大学,2008.
[23]曾俊.图像边缘检测技术及其应用研究[D].华中科技大学,2011.
[24]沈春裕.自适应阈值的运动目标检测技术研究[D].浙江工业大学,2012.
[25]蒋东升.基于数学形态学的边缘检测算法研究[D].电子科技大学,2012.
[26]Joshi Shantanu H,Srivastava Anuj. Intrinsic Bayesian Active Contours for Extraction of Object Boundaries in Images.[J]. International journal of computer vision,2009,81(3): 455-459.
[27]曾脉,左志宏.一种准确而快速的运动目标检测算法[J].成都信息工程学院学报,2008,23(4)：370-373.
[28]王建华.基于嵌入式系统的机器视觉及其在机器人控制中的应用[D].重庆：重庆大学硕士学位论文,2003.12.
[29]曲巨宝,林宏丛,梁洪涛,刘胜.运动图像快速跟踪技术研究[J].重庆师范大学学报,2011,28(1):44-48.
[30]张中方,李金屏,拜佩.基于多颜色空间融合的移动目标检测算法[J].济南大学学报(自然科学版),2011,(02)：191-195.
[31]赵欣,陈峰,吴立知.一种改进的meanshift运动目标跟踪算法[J].通信技术,2011,(11)：76-79.
[32]庞国瑞,葛广英,刘洪朋,肖新帅.基于改进的自适应混合高斯背景模型的车辆检测方法[J].聊城大学学报(自然科学版),2012,(01)：37-41.
[33]杨红霞,杭亦文,刘旭.基于Meanshift与Kalman的视频目标跟踪算法[J].武汉理工大学学报(信息与管理工程版),2012,(02)：147-150.
[34]王田,刘伟宁,韩广良.杜超,刘恋.基于改进MeanShift的目标跟踪算法[J].液晶与显示,2012,(03)：396-400.
[35]姚放吾,许辰铭.基于目标质心的Meanshift跟踪算法[J].计算机技术与发展,2012,(06)：104-106+110.
[36]卢官明,郎苏娟.基于YC bC r颜色空间的背景建模及运动目标检测[J].南京邮电大学学报(自然科学版),2009,06：17-22.
[37]徐晓军,纪玉波.基于HSV颜色空间静止背景中运动目标的检测算法[J].辽宁石油化工大学学报,2010,01：71-75.
[38]丁莹,钱锋,范静涛,姜会林.基于不同颜色空间的运动目标检测算法分析[J].长春理工 大学学报(自然科学版)，2012,04：1-4.
[39]徐以美,郭宝龙,陈龙.基于RGB颜色空间的减背景运动目标检测[J].计算机仿真,2008,03:214-217.
[40]沈盼盼,樊丰,伍瑞卿.基于RGB三通道分离的运动目标检测方法[J].电视技术,2012,36(3)：137-140.
[41]CHEN Jie,MA Tao,CHEN WenJie,PENG ZhiHong. Unsupervised robust recursive least-squares algorithm for impulsive noise filtering[J]. Science China (Information Sciences),2013,(04)203-212.
[42]元辉,李志斌,刘微.基于色度分割与图割算法的视差估计算法[J].华南理工大学学报(自然科学版),2013,(02)：12-18.
[43]雷禹.基于自适应背景替换的运动对象检测算法研究[D].广西大学,2012.
[44]艾凯文.基于自适应混合高斯模型的运动目标检测[D]广西大学,2012.
[45]秦臻,曹剑中. 基于方向直方图的Mean shift目标跟踪新算法[J].电子设计工程.2011,(12)：21-25.
[46]王宇翔,韩振铎,王宏敏,臧传吉.一种基于自适应背景预测的红外弱小目标检测方法[J].激光与红外,2012,(10)：1181-1186.
[47]胥欣,江登表,李勃,陈启美.混合高斯模型运动检测算法优化[J].计算机应用研究,2013.
[48]常向魁.视频运动目标跟踪算法研究[D].河南大学,2007.
[49]田莘.基于MeanShift算法的目标跟踪问题研究[D].西安科技大学,2010.
[50]亢朋朋,朱锡芳,南东亮,金运策.一种改进Mean shift算法的视频监控系统[J].微电子学与计算机,2013,02：153-156.