基于两标记点的大空间视频运动捕捉系统标定方法研究

The high-accuracy calibration of a video motion capture system is vital to the high-accuracy location of human posture. The commonly used method that first calibrates the intrinsic parameters of all cameras respectively and then performs global localization is complicated to implement, and the fabrication error of a 1D object with multiple markers will cause system error when it is used to perform simultaneous calibration of intrinsic and extrinsic parameters. A calibration object with two markers is prone to self-occlusion and easy to manufacture with a relatively large scale, which makes it suitable for the calibration of a large scale video motion capture system. This project researches the calibration method of a large scale video motion capture system with two markers, where the main contents include: transforming the original Euclidean upgrading constraint equations derived from the segment length prior, resulting in constraint equations with lower complexity of solving; constructing the general solving template for the transformed constraint equations in the real number field, and formulating a robust algorithm for the solving of the complex solution set of the Euclidean upgrading constraint equations; formulating the bundle adjustment algorithm that models the distance between sphere centers and the projection centers of spheres when spherical markers are used, where the accuracy of calibration result is close to the theoretical accuracy lower bound; designing the best structure parameters of the two spherical markers by modeling their influences to the calibration accuracy, and investigating how to measure the structure parameters with high accuracy. The research of this project is helpful to settle the robust computation of the Euclidean upgrading from segment lengths, and provide a high-accuracy and low-cost new method for calibrating a large scale video motion capture system.

视频运动捕捉系统的高精度标定对人体姿态的精确定位至关重要。先分别标定多摄像机内参数再统一定位的常用方法过程繁琐，而使用多标记点一维物体同时标定内外参数的方法受制作工艺影响存在系统误差。两标记点标定物不易自遮挡、易于制作较长尺寸，适用于大空间视频运动捕捉系统标定。本项目拟开展基于两标记点的大空间视频运动捕捉系统标定方法研究，主要研究内容包括：对基于线段长度的原欧氏升级约束方程组进行变换，建立求解复杂度更低的约束方程；在实数域内构造变换后约束方程的通用求解模板，构建欧氏升级约束方程复数解集的鲁棒算法；构建使用球形标记点时建模球心间距和球投影中心的捆绑调整算法，获得逼近理论精度下限的标定结果；通过建模两球形标记点的结构参数对标定精度的影响设计最优结构参数，并研究结构参数的高精度测量方法。本项目的研究有助于解决基于线段长度的鲁棒欧氏升级问题，为大空间视频运动捕捉系统标定提供高精度、低成本的新方法。

视频运动捕捉系统的高精度标定对人体姿态的精确定位至关重要。先分别标定多摄像机内参数再统一定位的常用方法过程繁琐，而使用多标记点一维物体同时标定内外参数的方法受制作工艺影响存在系统误差。两标记点标定物不易自遮挡、易于制作较长尺寸，适用于大空间视频运动捕捉系统标定。本项目开展了基于两标记点的大空间视频运动捕捉系统标定方法研究，主要研究内容包括：. （1）对基于线段长度的原欧氏升级约束方程组进行变换，建立求解复杂度更低的约束方程；. （2）在实数域内构造变换后约束方程的Gröbner基方法通用求解模板，构建欧氏升级约束方程复数解集的鲁棒算法，分析两球形标记点的结构参数对欧式升级和摄像机标定精度的影响；. （3）构建建模镜头畸变和部分标记点图像缺失的捆绑调整算法，获得逼近理论精度下限的标定结果；. （4）设计基于单摄像机和基于经纬仪的球体定位方法，用于测量自制两标记点标定物的球心距参数。. 在20米*40米的大空间视频运动捕捉场景中，基于两标记点标定物的多摄像机标定方法的目标定位平均误差小于1mm。基于单摄像机和基于经纬仪的球体定位方法在球心距测量时的误差小于2mm。本项目的研究有助于解决基于线段长度的鲁棒欧氏升级问题，为大空间视频运动捕捉系统标定提供高精度、低成本的新方法。