针对大视场范围下运动目标位姿参数测量易受模型累积误差、成像畸变和特征信息不足等因素的影响,提出一种新型视觉测量方法。首先建立适用于视觉测量过程中的高效多源特征数据融合模型,可以解决特征点单一的问题。然后构建基于特征点云信息的双向闭合测量模式,改变传统方法中从图像数据到空间特征信息的单向传递过程,将已确认的空间数据作为控制信息返回至测量处理过程中,可以有效规避测量空间越大导致测量模型累积误差越大的矛盾。最后实验结果表明,所提方法在10 m×8 m×3 m大视场空间内实现目标的姿态测量精度优于±1.5°,位置精度优于2 mm。所得的测量结果验证了双向闭合云控制测量模式获取的目标位姿参数精度高,稳定性强,能够满足实际的工程应用需求。

基于摄像头（可见光)的心率测量方法, 能够以非接触的方式检测受试者心率, 无论在临床应用还是在家庭健康监护中都有着极大的应用前景。 但是, CMOS摄像头的行扫描方式采集图像以及计算机系统的图像采集时钟抖动都会影响此类心率测量的精度, 从而引入相位误差和系统随机误差。 本文针对这两大主要误差进行研究分析, 提出了消除相位误差的基于傅里叶变换的幅频叠加算法、 消除图像采集系统时钟抖动误差的基于时间标定的三次样条插值重构方法。 幅频叠加算法只在幅频域对信号进行处理, 能够忽略信号的相位影响, 从而消除相位误差。 时间标定的三次样条重构算法能够重构图像的均匀采集, 从而消除系统时钟抖动引起的随机误差。 同时, 通过理论推导论证了两种算法的可行性, 进而利用LED模拟实验和实际心率测量实验验证该算法在提高信号检测精密度中的特性。 在模拟实验中, 对每帧图像的200行作幅频叠加运算, 能够使信号幅值相对提升458%; 基于时间标定的三次样条重构算法能使检测的信号频率的均方根误差缩小30%以上; 在实际心率检测中, 幅频叠加算法可使心率幅值相对提升达335%, 基于时间标定的三次样条重构算法可使心率准确度提升40%左右。 因此, 模拟实验和实际心率测量实验均验证了提出的算法在提高系统检测精度上的有效性, 提高了基于机器视觉心率测量的抗干扰能力。 这两种算法不仅能够提高基于机器视觉的心率检测精度, 而且能够适用于基于机器视觉对一定频率信号的检测, 在机器视觉检测中具有广泛的意义。

We propose an approach for realizing 3D reconstruction in small field of view or some extreme environments. We combine the stereo vision and the structured light technologies and change the traditional ways of ap-plying them by employing two fiber bundles. The processes of calibration and 3D reconstruction are also introduced. Experiments are performed to verify the feasibility and effectiveness of our proposed approach.

A two-step method for pose estimation based on five co-planar reference points is studied. In the first step, the pose of the object is estimated by a simple analytical solving process. The pixel coordinates of reference points on the image plane are extracted through image processing. Then, using affine invariants of the reference points with certain distances between each other, the coordinates of reference points in the camera coordinate system are solved. In the second step, the results obtained in the first step are used as initial values of an iterative solving process for gathering the exact solution. In such a solution, an unconstrained nonlinear optimization objective function is established through the objective functions produced by the depth estimation and the co-planarity of the five reference points to ensure the accuracy and convergence rate of the non-linear algorithm. The Levenberg-Marquardt optimization method is utilized to refine the initial values. The coordinates of the reference points in the camera coordinate system are obtained and transformed into the pose of the object. Experimental results show that the RMS of the azimuth angle reaches 0.076o in the measurement range of 0o-90o; the root mean square (RMS) of the pitch angle reaches 0.035o in the measurement range of 0o-60o; and the RMS of the roll angle reaches 0.036o in the measurement range of 0o-60o.

The pose estimation method based on geometric constraints is studied. The coordinates of the five feature points in the camera coordinate system are calculated to obtain the pose of an object on the basis of the geometric constraints formed by the connective lines of the feature points and the coordinates of the feature points on the CCD image plane; during the solution process, the scaling and orthography projection model is used to approximate the perspective projection model. The initial values of the coordinates of the five feature points in the camera coordinate system are obtained to ensure the accuracy and convergence rate of the non-linear algorithm. In accordance with the perspective projection characteristics of the circular feature landmarks, we propose an approach that enables the iterative acquisition of accurate target poses through the correction of the perspective projection coordinates of the circular feature landmark centers. Experimental results show that the translation positioning accuracy reaches ±0.05 mm in the measurement range of 0–40 mm, and the rotation positioning accuracy reaches ±0.06o in the measurement range of 4o–60o.

Based on structured-light vision measurement technology, we study a measuring method for microdiameter. The measurement principle and mathematical model are described. A novel grayscale barycenter extraction algorithm along the radial direction is proposed, which can precisely gather the image coordinates of the ellipse-shaped light-stripe centers. The accuracy of the measurement result shows marked improvement by using the algorithm. The method executes circle fitting to the measured three-dimensional (3D) data using linear least square method, which can acquire the diameter, surface profile, and other information of the object effectively. On the scene, a line-structured light vision system using the presented method is applied to measure the curvature radius of metal blades. Experimental results show that the measurement precision of the system is higher than 2 \mu m.

In order to estimate the position and orientation of an object with a single camera, a novel measurement method based on pinhole camera model with five reference points is presented. Taking the specially designed planar target with the monocular vision system, the projection line of the reference points is built. According to the projection model, the coordinates of the reference points in the camera coordinate system are estimated with the least-squares algorithm. Thus the position and orientation of the target are worked out. Experimental result shows that the measurement precision of angle is less than 0.2°, and that of displacement is less than 0.1 mm.

提出一种三维物体360°面形自动测量方法.简要介绍了方法原理,系统结构,信息获取与处理方案,并给出对圆锥-圆台金属模块和人头石膏模型360°面形测量结果.