移动机器人的目标识别研究
|
摘要
移动机器人的目标识别是以图像处理、分析和理解为基础,是一项多学科综合的复杂技术,现在已经渗透到军事、空间探索、医学、工业等各个领域,具有很大的适用价值和重要意义。本文是对移动机器人的目标识别技术进行研究,主要完成以下几个方面的工作:
首先搭建移动机器人的软硬件平台,采用上下位机的结构,下位机采用于S3C2410和C8051F350作为处理器并搭建必要外部电路,用于对图像信息的采集及机器人自主移动的控制。下位机采集图像信息后,通过无线网络将图像数据实时传送至上位机,上位机选用PC机,用于对图像数据进行预处理及目标识别,最后将识别结果传回移动机器人作为机器人的决策依据。
其次就是分析移动机器人的主要噪声干扰,介绍在室内环境下的图像滤波算法和二值化算法,通过实验仿真,选择对抖动噪声具有抑制作用的滤波算法,提出Ostu阈值分割算法与Sobel边缘检测算子相结合的图像二值化处理策略。
接着研究对预处理后的目标图像的特征提取,在分析相对不变矩、仿射不变矩、Zernike矩以及NMI优缺点的基础上,提出了一种新的组合不变矩,并对其进行重点研究,通过实验仿真表明,新提出的组合不变矩不仅具有类内聚合类间可分性,而且还具有一定的抗干扰能力,可以用来作为移动机器人目标识别的特征向量。
最后介绍RBF神经网络模式分类器,提出RBF神经网络的训练算法,并提出应用RBF神经网络完成基于图像组合不变矩的目标识别算法,最后通过对均匀性与非均匀性背景的相似目标图像进行实验仿真表明,该算法能准确的识别出目标对象。
Target recognition of mobile robot is based on image processing, analysis and understanding. It is a multidisciplinary integrated complex technology, and it has been applied successfully in the military, space exploration, medical and industrial fields, etc. So it has greatly importance and practical value. In this paper, the main works which has been done are as following:
Firstly, the software and hardware platform of mobile robot ware been built, the structure of the system adopted two host, one host was C8051F350 processor ,S3C2410 processor based on ARM920T and essential hardware, which is used for image acquisition and controlling the motion of mobile robot. When the host acquisitioned image, Then transfer the collected target image to PC by wireless LAN in real time. The other host is PC, which is used for pre-processing and recognizing target image which is received from S3C2410. After completing the image recognition on PC, then return the recognition results to the mobile robot.
Secondly, Analyses the main noise of mobile robot and research the image filtering algorithm and binary algorithm in indoor environment. Select the filtering algorithm for the jitter noise, Image binarization strategy which is Ostu threshold segmentation algorithm combining with Sobel edge detection was proposed via experiment and simulation.
Thirdly, the research is carried out on the feature extraction of the target image after pre-processing. A new kind of combined invariant moments is proposed and researched here based on contrast of the relatively constant moments, affine invariant moments, Zernike moments and NMI. Simulation experiment shows that the new kind of combined moments has polymerizability within the class and has separability without the class and at the same time has a certain anti-interference ability, can be used as the feature vector for target recognization on the mobile robot.
Finally, the paper introduced the RBF neural network model, puts forward the RBF neural network classifier training algorithm, and proposed a target recognition algorithm which is based on combining RBF neural network and combined invariant moments. Finally, through simulation experiment on the similar target image with uniform and non-uniform background, the algorithm proves to identify the target accurately. Finally, through simulation experiment on the similar target image with uniform and non-uniform background, the algorithm proves to identify the target accurately.
首先搭建移动机器人的软硬件平台,采用上下位机的结构,下位机采用于S3C2410和C8051F350作为处理器并搭建必要外部电路,用于对图像信息的采集及机器人自主移动的控制。下位机采集图像信息后,通过无线网络将图像数据实时传送至上位机,上位机选用PC机,用于对图像数据进行预处理及目标识别,最后将识别结果传回移动机器人作为机器人的决策依据。
其次就是分析移动机器人的主要噪声干扰,介绍在室内环境下的图像滤波算法和二值化算法,通过实验仿真,选择对抖动噪声具有抑制作用的滤波算法,提出Ostu阈值分割算法与Sobel边缘检测算子相结合的图像二值化处理策略。
接着研究对预处理后的目标图像的特征提取,在分析相对不变矩、仿射不变矩、Zernike矩以及NMI优缺点的基础上,提出了一种新的组合不变矩,并对其进行重点研究,通过实验仿真表明,新提出的组合不变矩不仅具有类内聚合类间可分性,而且还具有一定的抗干扰能力,可以用来作为移动机器人目标识别的特征向量。
最后介绍RBF神经网络模式分类器,提出RBF神经网络的训练算法,并提出应用RBF神经网络完成基于图像组合不变矩的目标识别算法,最后通过对均匀性与非均匀性背景的相似目标图像进行实验仿真表明,该算法能准确的识别出目标对象。
Target recognition of mobile robot is based on image processing, analysis and understanding. It is a multidisciplinary integrated complex technology, and it has been applied successfully in the military, space exploration, medical and industrial fields, etc. So it has greatly importance and practical value. In this paper, the main works which has been done are as following:
Firstly, the software and hardware platform of mobile robot ware been built, the structure of the system adopted two host, one host was C8051F350 processor ,S3C2410 processor based on ARM920T and essential hardware, which is used for image acquisition and controlling the motion of mobile robot. When the host acquisitioned image, Then transfer the collected target image to PC by wireless LAN in real time. The other host is PC, which is used for pre-processing and recognizing target image which is received from S3C2410. After completing the image recognition on PC, then return the recognition results to the mobile robot.
Secondly, Analyses the main noise of mobile robot and research the image filtering algorithm and binary algorithm in indoor environment. Select the filtering algorithm for the jitter noise, Image binarization strategy which is Ostu threshold segmentation algorithm combining with Sobel edge detection was proposed via experiment and simulation.
Thirdly, the research is carried out on the feature extraction of the target image after pre-processing. A new kind of combined invariant moments is proposed and researched here based on contrast of the relatively constant moments, affine invariant moments, Zernike moments and NMI. Simulation experiment shows that the new kind of combined moments has polymerizability within the class and has separability without the class and at the same time has a certain anti-interference ability, can be used as the feature vector for target recognization on the mobile robot.
Finally, the paper introduced the RBF neural network model, puts forward the RBF neural network classifier training algorithm, and proposed a target recognition algorithm which is based on combining RBF neural network and combined invariant moments. Finally, through simulation experiment on the similar target image with uniform and non-uniform background, the algorithm proves to identify the target accurately. Finally, through simulation experiment on the similar target image with uniform and non-uniform background, the algorithm proves to identify the target accurately.
引文
[1]夏庭锴,杨明,杨汝清.基于单目视觉的移动机器人导航算法研究进展[J].控制与决策,2010,25(1).
[2]赵金英张铁中杨丽.西红柿采摘机器人视觉系统的目标提取[J].农业机械学报,2006,36(10):200-203.
[3] Lie Guo, Linhui Li, Yibing Zhao, Mingheng Zhang. Study on Pedestrian Detection and Tracking with Monocular Vision[C]. 2010 2nd International Conference on Computer Technology and Development (ICCTD 2010).
[4]郭亚.汽车牌照识别系统中的牌照定位方法研究[D].陕西:长安大学,2008.
[5]任明武,杨静宇,孙涵。一种基于边缘模式的直方图构造新方法[J].计算机研究与发展,2001,38(8):972-976.
[6]赵雪松,陈淑珍.综合全局二值比与边缘检测的图像分割方法[J].计算机辅助设计与图形学学报,2001,13(2):118-121.
[7]金立左,夏良正.图像分割的自适应魔术阈值法[J].中国图像图形学报,2000,5(5):390-395
[8] CHEN Zi-Kuan, TAO Yang, CHEN Xin, etal. Wavelet Based Adaptive Threshloding Method for Image Segmentation [J].Optical Enineering, 2001, 40(5):868-874.
[9]张斌,朱正中.基于边缘轮廓信息的多源遥感图像分割[J].中国图像图形学报,2003,3(5):830-835.
[10]周凌翔,顾伟康.最佳边缘检测的准则与算子[J].模式识别与人工智,1998,11(1):54-61.
[11]王广君,田金文,柳健.基于四叉树结构的图像分割技术[J].华中科技大学学报,2000,(2):12-16.
[12] LORENZ Thmos. Set_Valued Mapsfor Image Segmentation[J]. Computing and Visualization,2001(5):45-57.
[13]刘极峰易际明.机器人技术基础[M].北京:高等教育出版社,2006.
[14]陈赜,秦贵和,徐华中,王磊等.ARM9嵌入式技术及Linux高级实践教程[M].北京:北京航空航天大学出版社,2005.
[15] webey官方网站http://www.webeye.com.cn/.
[16] M22 GSM/GPRS数据手册(第一版),台湾BENQ公司.
[17]袁俊杰.基于ARM9的嵌入式移动机器人远程控制的研究[D].天津:天津理工大学
[18]曾万梅.基于图像的目标快速识别与跟踪技术研究[D].南京:南京航空航天大学
[19] N.Otsu. A threshold selection method from gray-level histograms. IEEE Trans Systems,Man and Cybernetics, 1979, 9(1):62-66.
[20] R.Guo, S.M. Pandit. Automatic threshold selection based on histogram modes and a Discriminant criterion[J]. Machine Vision and Application, 1998, 10:331~338.
[21]夏良正,李久贤编著。数字图像处理[M].东南大学出版社.
[22]张毅军,吴雪箐,夏良正.二维熵图像阈值分割的递推快速算法[J].模式识别与人工智能.1997,10(3).
[23]王茜,彭中,刘莉.一种基于自适应阈值的图像分割算法[J].北京理工大学学报,2003,23(4):521-524.
[24] canny John. A Computatuinal Approach to Edge Detection[J].IEEE Transactins on Pattern Analysis and Machine Intelligence, 1986, PAMI-8(6) 679-698
[25] Maria Petrou(英),Panagiota Bosdogianni(希)著,赖剑煌,冯国灿等译。数字图像处理疑难解析[M]。机械工业出版社,2004.
[26]赵荣椿,赵忠明等.数字图像处理导论[M].西安:西北工业大学出版社,2000,194-195.
[27] Hu.M.Visual pattern recognition by moment invariant[J].IRE Trans- on Inf.Theory,1986:179-187.
[28]丁明跃,常金玲,彭嘉熊.不变矩算法研究[J].数据采集与处理,1992,7(1).
[29]王其聪.基于小波分析的矩特征和神经网络的图像识别[D].浙江工业大学.
[30] Chen C C. Improved moment invadants for shape discrimination[J]. Pattern Recognition. V01.26.No.5.1993.
[31]王波涛,孙景鳌,蔡安妮.相对矩及在几何形状识别中的应用[J].中国图象图形学报,2001,6(3):296-300.
[32] A Sluzek.Identification and inspection of 2-D objects using new moment based shape descriptors[J].Pattern Recognition,V01.16.1995:687—697.
[33]李迎春,陈贺新,赵铭等。基于飞机目标的仿射不变矩研究[J].吉林大学学报(信息科学版).2003,21:84-88.
[34] M.R. Teague. Image analysis via the general theory of moments[J]. Optical Soc. Am. 70 (1980) 920-930.
[35] Sun-Kyoo Hwang,Whoi-Yul Kim。Anovel approach to the fast computation of Zernike moments[J].Pattern Recognition,2006.
[36] Chandan Singh,Ekta Walia. Algorithms for fast computation of Zernike moments and their numerical stability[J],Pattern Recognition,2011
[37]潘昊,田捷,钟珞.前馈多层神经网络的步长搜索调整研究[J].计算机工程与就用,2004,(4):17-18.
[38]饶文碧,吴代华.RBF神经网络及其在结构损伤识别中的应用与研究[J].固体力学学报2002,23(4):477-482.
[2]赵金英张铁中杨丽.西红柿采摘机器人视觉系统的目标提取[J].农业机械学报,2006,36(10):200-203.
[3] Lie Guo, Linhui Li, Yibing Zhao, Mingheng Zhang. Study on Pedestrian Detection and Tracking with Monocular Vision[C]. 2010 2nd International Conference on Computer Technology and Development (ICCTD 2010).
[4]郭亚.汽车牌照识别系统中的牌照定位方法研究[D].陕西:长安大学,2008.
[5]任明武,杨静宇,孙涵。一种基于边缘模式的直方图构造新方法[J].计算机研究与发展,2001,38(8):972-976.
[6]赵雪松,陈淑珍.综合全局二值比与边缘检测的图像分割方法[J].计算机辅助设计与图形学学报,2001,13(2):118-121.
[7]金立左,夏良正.图像分割的自适应魔术阈值法[J].中国图像图形学报,2000,5(5):390-395
[8] CHEN Zi-Kuan, TAO Yang, CHEN Xin, etal. Wavelet Based Adaptive Threshloding Method for Image Segmentation [J].Optical Enineering, 2001, 40(5):868-874.
[9]张斌,朱正中.基于边缘轮廓信息的多源遥感图像分割[J].中国图像图形学报,2003,3(5):830-835.
[10]周凌翔,顾伟康.最佳边缘检测的准则与算子[J].模式识别与人工智,1998,11(1):54-61.
[11]王广君,田金文,柳健.基于四叉树结构的图像分割技术[J].华中科技大学学报,2000,(2):12-16.
[12] LORENZ Thmos. Set_Valued Mapsfor Image Segmentation[J]. Computing and Visualization,2001(5):45-57.
[13]刘极峰易际明.机器人技术基础[M].北京:高等教育出版社,2006.
[14]陈赜,秦贵和,徐华中,王磊等.ARM9嵌入式技术及Linux高级实践教程[M].北京:北京航空航天大学出版社,2005.
[15] webey官方网站http://www.webeye.com.cn/.
[16] M22 GSM/GPRS数据手册(第一版),台湾BENQ公司.
[17]袁俊杰.基于ARM9的嵌入式移动机器人远程控制的研究[D].天津:天津理工大学
[18]曾万梅.基于图像的目标快速识别与跟踪技术研究[D].南京:南京航空航天大学
[19] N.Otsu. A threshold selection method from gray-level histograms. IEEE Trans Systems,Man and Cybernetics, 1979, 9(1):62-66.
[20] R.Guo, S.M. Pandit. Automatic threshold selection based on histogram modes and a Discriminant criterion[J]. Machine Vision and Application, 1998, 10:331~338.
[21]夏良正,李久贤编著。数字图像处理[M].东南大学出版社.
[22]张毅军,吴雪箐,夏良正.二维熵图像阈值分割的递推快速算法[J].模式识别与人工智能.1997,10(3).
[23]王茜,彭中,刘莉.一种基于自适应阈值的图像分割算法[J].北京理工大学学报,2003,23(4):521-524.
[24] canny John. A Computatuinal Approach to Edge Detection[J].IEEE Transactins on Pattern Analysis and Machine Intelligence, 1986, PAMI-8(6) 679-698
[25] Maria Petrou(英),Panagiota Bosdogianni(希)著,赖剑煌,冯国灿等译。数字图像处理疑难解析[M]。机械工业出版社,2004.
[26]赵荣椿,赵忠明等.数字图像处理导论[M].西安:西北工业大学出版社,2000,194-195.
[27] Hu.M.Visual pattern recognition by moment invariant[J].IRE Trans- on Inf.Theory,1986:179-187.
[28]丁明跃,常金玲,彭嘉熊.不变矩算法研究[J].数据采集与处理,1992,7(1).
[29]王其聪.基于小波分析的矩特征和神经网络的图像识别[D].浙江工业大学.
[30] Chen C C. Improved moment invadants for shape discrimination[J]. Pattern Recognition. V01.26.No.5.1993.
[31]王波涛,孙景鳌,蔡安妮.相对矩及在几何形状识别中的应用[J].中国图象图形学报,2001,6(3):296-300.
[32] A Sluzek.Identification and inspection of 2-D objects using new moment based shape descriptors[J].Pattern Recognition,V01.16.1995:687—697.
[33]李迎春,陈贺新,赵铭等。基于飞机目标的仿射不变矩研究[J].吉林大学学报(信息科学版).2003,21:84-88.
[34] M.R. Teague. Image analysis via the general theory of moments[J]. Optical Soc. Am. 70 (1980) 920-930.
[35] Sun-Kyoo Hwang,Whoi-Yul Kim。Anovel approach to the fast computation of Zernike moments[J].Pattern Recognition,2006.
[36] Chandan Singh,Ekta Walia. Algorithms for fast computation of Zernike moments and their numerical stability[J],Pattern Recognition,2011
[37]潘昊,田捷,钟珞.前馈多层神经网络的步长搜索调整研究[J].计算机工程与就用,2004,(4):17-18.
[38]饶文碧,吴代华.RBF神经网络及其在结构损伤识别中的应用与研究[J].固体力学学报2002,23(4):477-482.