嗅觉搜索机器人的单目视觉室内定位技术研究
|
摘要
针对室内嗅觉搜索机器人定位问题,提出一种基于最小二乘高斯核支持向量机的单目视觉定位算法.首先,利用最小二乘高斯核支持向量机,可将机器人在图片中像素位置与真实场景中位置间的非线性映射转化为带高斯核的线性映射;然后,运用k重交叉验证方法对映射关系模型进行训练,获取优化模型参数;最后,利用单目摄像头采集机器人运动的连续图像,并根据机器人平台与其背景间的灰度差异特征实时识别图像中目标机器人平台,计算得到机器人平台形心在每一帧图像中的像素位置,进而映射出机器人平台在室内环境下的实时动态位置.实验结果表明:此定位方法能够为移动机器人在室内环境下提供高精度和无累积误差的实时定位信息;算法的精确性和可靠性得到了充分验证.
In order to localize olfactory-based searching mobile robots in the indoor environments,a monocular-vision localization algorithm was proposed based on least square support vector machine(LS-SVM) method.By using Gaussian kernel LS-SVM classifier,the non-linear mapping between robot real positions to pixels in the image could be translated into linear mapping with Gaussian kernel.Then,k-fold cross validation was used to seek the optimized parameters after linear mapping model training.Finally,continuously moving robot was identified based on greyscale level in real time from sequential images reordered by a single camera.The real-time position of robot in the indoor environment was calculated from the vehicles' geometric center pixel on every image using the linear mapping model developed herein.Experiment results show that the proposed localization algorithm can provide high precision real-time position information for robots in the indoor environment without error accumulated with time.The accuracy and reliability of the algorithm were fully verified.
In order to localize olfactory-based searching mobile robots in the indoor environments,a monocular-vision localization algorithm was proposed based on least square support vector machine(LS-SVM) method.By using Gaussian kernel LS-SVM classifier,the non-linear mapping between robot real positions to pixels in the image could be translated into linear mapping with Gaussian kernel.Then,k-fold cross validation was used to seek the optimized parameters after linear mapping model training.Finally,continuously moving robot was identified based on greyscale level in real time from sequential images reordered by a single camera.The real-time position of robot in the indoor environment was calculated from the vehicles' geometric center pixel on every image using the linear mapping model developed herein.Experiment results show that the proposed localization algorithm can provide high precision real-time position information for robots in the indoor environment without error accumulated with time.The accuracy and reliability of the algorithm were fully verified.
引文
[1]李群明,熊蓉,褚健.室内自主移动机器人定位方法研究综述[J].机器人,2003,25(6):560-567.
[2]王殿君,魏洪兴,任福君.移动机器人自主定位技术[M].北京:机械工业出版社,2013.
[3]韩冲,苏涛,谢基榕.基于OpenCV的水下机器人单目定位技术研究与仿真[J].计算机测量与控制,2017(12):219-223.
[4]LI Y,JIANG Y,CAO J,et al.AUV docking experiments based on vision positioning using two cameras[J].Ocean Engineering,2015,110:163-173.
[5]崔宝侠,张驰,栾婷婷,等.基于全景视觉机器人定位的路标提取方法[J].沈阳工业大学学报,2016,38(5):526-530.
[6]张怀相,朱金成.基于双目视觉的避障策略算法研究[J].杭州电子科技大学学报,2013(4):31-34.
[7]贺勇标,杨天龙,鲁国峰,等.基于线结构光的双目视觉目标位姿测量研究[J].机械工程师,2017(5):1-3.
[8]PRIYANTHA N,CHAKRABORTY A,BALAKRISH-NAN H.The cricket location-support system[C]//Proc of The 6th ACM International Conference on Mobile Computing and Networking.Boston:ACM,2000:32-43.
[9]李帅鑫,李广云,周阳林,等.改进的单目视觉实时定位与测图方法[J].仪器仪表学报,2017(11):2849-2857.
[10]李繤,刘瑶,洪良,等.室内单目视觉全局定位系统的设计与实现[J].工业仪表与自动化装置,2017(1):13-19.
[11]苑晶,刘钢墩,孙沁璇.激光与单目视觉融合的移动机器人运动目标跟踪[J].控制理论与应用,2016,33(2):196-204.
[12]梁潇.基于激光与单目视觉融合的机器人室内定位与制图研究[D].哈尔滨:哈尔滨工业大学图书馆,2016.
[13]AFIA A B,DEAMBROGIO L,SALOS D,et al.Review and classification of vision-based localisation techniques in unknown environments[J].Radar Sonar&Navigation Iet,2014,8(9):1059-1072.
[14]BRADSKI G.The opencv library[J].Doctor Dobbs Journal,2000,25(11):384-386.
[15]VAPNIK V,CHAPELLE O.Bounds on error expectation for support vector machines[J].Neural Computation,2000,12(9):2013-2036.
[16]SUYKENS J A K,VANDEWALLE J.Least squares support vector machine classifiers[M].Dordrecht:Kluwer Academic Publishers,1999.
[2]王殿君,魏洪兴,任福君.移动机器人自主定位技术[M].北京:机械工业出版社,2013.
[3]韩冲,苏涛,谢基榕.基于OpenCV的水下机器人单目定位技术研究与仿真[J].计算机测量与控制,2017(12):219-223.
[4]LI Y,JIANG Y,CAO J,et al.AUV docking experiments based on vision positioning using two cameras[J].Ocean Engineering,2015,110:163-173.
[5]崔宝侠,张驰,栾婷婷,等.基于全景视觉机器人定位的路标提取方法[J].沈阳工业大学学报,2016,38(5):526-530.
[6]张怀相,朱金成.基于双目视觉的避障策略算法研究[J].杭州电子科技大学学报,2013(4):31-34.
[7]贺勇标,杨天龙,鲁国峰,等.基于线结构光的双目视觉目标位姿测量研究[J].机械工程师,2017(5):1-3.
[8]PRIYANTHA N,CHAKRABORTY A,BALAKRISH-NAN H.The cricket location-support system[C]//Proc of The 6th ACM International Conference on Mobile Computing and Networking.Boston:ACM,2000:32-43.
[9]李帅鑫,李广云,周阳林,等.改进的单目视觉实时定位与测图方法[J].仪器仪表学报,2017(11):2849-2857.
[10]李繤,刘瑶,洪良,等.室内单目视觉全局定位系统的设计与实现[J].工业仪表与自动化装置,2017(1):13-19.
[11]苑晶,刘钢墩,孙沁璇.激光与单目视觉融合的移动机器人运动目标跟踪[J].控制理论与应用,2016,33(2):196-204.
[12]梁潇.基于激光与单目视觉融合的机器人室内定位与制图研究[D].哈尔滨:哈尔滨工业大学图书馆,2016.
[13]AFIA A B,DEAMBROGIO L,SALOS D,et al.Review and classification of vision-based localisation techniques in unknown environments[J].Radar Sonar&Navigation Iet,2014,8(9):1059-1072.
[14]BRADSKI G.The opencv library[J].Doctor Dobbs Journal,2000,25(11):384-386.
[15]VAPNIK V,CHAPELLE O.Bounds on error expectation for support vector machines[J].Neural Computation,2000,12(9):2013-2036.
[16]SUYKENS J A K,VANDEWALLE J.Least squares support vector machine classifiers[M].Dordrecht:Kluwer Academic Publishers,1999.