基于单目视觉的特征点测距与定位方法研究  

Feature Point Ranging and Positioning Method Based on Monocular Vision

作　　者：承达瑜 张新 孟小亮 宿文松 莫新好 任明龙 王腾 曾广鸿 CHENG Dayu;ZHANG Xin;MENG Xiaoliang;SU Wensong;MO Xinhao;REN Minglong;WANG Teng;ZENG Guanghong(School of Mining and Geomatics Engineering,Hebei University of Engineering,Handan 056038,China;Research Center of Hebei Province Ecological Civilization and Social Governance,Hebei University of Engineering,Handan 056038,China;School of Remote Sensing Information Engineering,Wuhan University,Wuhan 430079,China;Guangdong Feida Traffic Engineering Co.,Ltd.,Guangzhou 510030,China;Guangzhou Expressway Co.,Ltd.,Guangzhou 510290,China;Surveying and Mapping Institute Lands and Resource Department of Guangdong Province,Guangzhou 510663,China;Guangdong Huitu Technology Development Co.,Ltd.,Guangzhou 510600,China)

机构地区：[1]河北工程大学矿业与测绘工程学院,河北邯郸056038 [2]河北工程大学河北省水生态文明及社会治理中心,河北邯郸056038 [3]武汉大学遥感信息工程学院,湖北武汉430079 [4]广东飞达交通工程有限公司,广东广州510030 [5]广州市高速公路有限公司,广东广州510290 [6]广东省国土资源测绘院,广东广州510663 [7]广东慧图资环科技发展有限公司,广东广州510600

出　　处：《测绘地理信息》2024年第2期25-32,共8页Journal of Geomatics

基　　金：国家重点研发计划(2020YFD1100203);河北省重大科技成果转化专项(22287401Z);科技基础资源调查专项(2019FY202503)。

摘　　要：在单目视觉测距系统中,摄像机安装偏差会造成摄像机云台俯仰角与模型俯仰角不一致,从而影响远距离单目测距和定位的精度。针对这一问题,提出了一种基于梯度下降法的俯仰角校正方法。该方法根据单目测距原理和参考点与摄像机的实际距离,构建出梯度下降法的目标函数;对目标函数进行一阶微分,求出目标函数的梯度;迭代计算出使目标函数值小于设定阈值的俯仰角,以此为单目测距模型最优俯仰角。在此基础上,依据单目视觉成像原理构建像素坐标与实际坐标间的函数关系,计算特征点的实际坐标。以球型摄像机为图像采集设备开展实验与应用,结果表明,测距结果的相对误差都在4%以内。这证明该方法能准确校正单目测距俯仰角,确保测距精度,且能有效确定特征点坐标。In the monocular vision ranging system,the camera installation deviation will cause the inconsistencies between the camera head's pitch angle and the model's pitch angle,which will affect the accuracy of the long-distance monocular vision ranging and positioning.To solve this problem,we propose a pitch angle correction method based on gradient descent method.According to the principle of monocular vision ranging and the actual distance between the reference point and the camera,the objective function of the gradient descent method is constructed.By differentiating the first order of the objective function,the gradient of the objective function is obtained.The pitch angle that makes the value of the objective function less than the set threshold is calculated iteratively,which is taken as the optimal pitch angle of the monocular vision ranging model.Under these conditions,the function relationship between pixel coordinates and actual coordinates is constructed according to the monocular vision imaging principle,and the actual coordinates of feature points are calculated.The experiment and application of spherical camera as image acquisition equipment show that the relative errors of ranging results are within 4%,which proves that the method can accurately correct the pitch angle of monocular vision ranging,ensure the ranging accuracy,and effectively determine the coordinates of feature points.

关 键 词：计算机视觉 单目测距 俯仰角 梯度下降法 球型摄像机 

分 类 号：P258[天文地球—测绘科学与技术] TP391[自动化与计算机技术—计算机应用技术]