显微图像轮廓提取的基元兴趣提取方法及其在姿态测量中的应用

1348 IEEE TRANSACTIONS ON SYSTEMS, MAN, AND CYBERNETICS: SYSTEMS, VOL. 48, NO. 8, AUGUST 2018

Contour Primitives of Interest Extraction

Method for Microscopic Images and Its

Application on Pose Measurement

Fangbo Qin, Fei Shen, Dapeng Zhang, Xilong Liu, and De Xu, Senior Member, IEEE

Abstract—This paper proposes a suite of methods to real-

ize high precision pose measurement in 3-D Cartesian space

based on a multicamera microscopic vision system. Since it is

inefficient to develop a specific image algorithm for each kind

along the normal vectors of these contour primitives. The posi-

tions and directional derivatives of these edges are used for

feature extraction and autofocus, respectively. The point fea-

tures and line features extracted from multiview images are

utilized to measure 3-D vectors and orientations, respectively,

based on image Jacobian matrices. Cameras’ linear motions

are considered in the imaging model, so that the measurement

range is expanded beyond the limitation of microscopes’ shal-

low depths of field. The affine epipolar constraint and focused

planes intersection constraint between cameras are applied to

in orientation, while the

measurement range is about 5000 µm in position and 20

◦

orientation.

Index Terms—Geometric constraint, image feature extraction,

microscopic vision, pose measurement, precision assembly

M

paper was recommended by Associate Editor Z. Liu.

Institute of Automation, Chinese Academy of Sciences, Beijing 100190,

University of Chinese Academy of Sciences, Beijing 101408, China (e-mail:

2168-2216

c

 2017 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.

QIN et al.: CPIE METHOD FOR MICROSCOPIC IMAGES AND ITS APPLICATION ON POSE MEASUREMENT 1349

and algorithms need be fast for real time operation.

Autofocusing is necessary to a microscopic camera, which

realizes acquiring clear images by adjusting its object dis-

tance automatically. It is implemented by translating the

camera along a linear motion stage to search for the posi-

tion where the focus measure (FM) is maximum. The FM

versus camera position curve is expected to have a sharp

peaks at other positions. The mountain climbing algorithm is

a broadly adopted searching strategy with high efficiency [11].

As is different from conventional lens, microscope is

with the limitations of narrow field of view and shal-

low depth of field, which are challenges to microscopic

vision. Monocular microscopic camera is suitable for planar

measurement [13]–[15]. Wang et al. [16] presented an auto-

mated 3-D micro-grasping method, in which the microscopic

camera offered 2-D position feedback. 3-D information of

object could be recovered from multiview images using stereo

vision [17], [18]. Yamamoto and Sano [19] used a stereo-

scopic microscope to measure the needle tip’s 3-D position

on visual feedbacks of multiple microscopic cameras in the

3-D precision assembly tasks. The objects’ relative poses were

measured using image Jacobian matrices and image features.

However, their relative position measurements were limited

within the optical depth of field, i.e., submillimeter range. In

many cases, the relative depth between two objects exceeds

the optical depth of field, so that the camera needs to be

translated to focus on them sequentially. The measurement