| Determining of the objects’ position and orientation(pose) has a wide range of applications in the fields of large-scale equipment manufacturing, such as aerospace, aircraft, shipbuilding, etc. With the development of the components assembling putting forward more and more high demand, the unstressed accurate digital assembly and flexible automatic assembly have attracted many reseachers’ interest. Measurement Assisted Assambly(MAA), using field measure to obtain the real-time data of assembly pose and feedback to guide the assembly, becomes more and more important. This study focuses on key technologies of vision-based pose measurement for large-scale equipment assembly, and the contributions are in 5 ways:1. A novel active defocused imaging technology and adaptive control for the light intensity of IR-LED target spots in large range are proposed and performed. In the pose measurement of large-scale equipment based on vision theory in a wide measurement space, it is a problem that at close distance the light adjustable resolution of IR-LEDs is not high enough, and at far distance the light intensity is low. In order to solve this problem, a technique of Pulse Width Modulation(PWM) with Multi-level Voltage Automatic Switch(MVAS) is proposed, and achieves an adaptive control for the IR-LED light intensity. Aim at the problem of the the limited depth and range by focusing target points to image, based on the analysis of imaging and diffraction principle and the experiment induction, the shape and size of a aperture and other imaging parameters are rationally designed and set, then the initiative defocused imaging IR-LEDs is achieved. Thus, the target points at different distance in the range of 3~12 m all can be imaged to high-quality spots.2. Based on the deep studies of target point imaging principle and the Point Spread Function(PSF) of the system, a novel subpixel centroid determining method based on a free-form surface is presented. For the large-scale equipment pose measurement, the IR-LED target point can be regarded as point light source, of which the spots have the features of small range and no obvious edge. The proposed method locates the target spots based on the spot gray distribution, which conforms to the trait of the spots. Because of defocusing the target points to image, the point spread function of optical imaging system is the composite of defocus PSF and Gauss PSF. Using free-form surface fitting method, not only avoids the complex computations of compound function problem, but also more accurately describes the gray distribution of target spot, and the accurate determining target spot centroid can be realized.3. In order to calibrate camera for a wide area vision measurement, a novel calibration method based on multiple direction virtual stereo reference is proposed. In this method, a virtual stereo reference is formed by precisely controlly single IR-LED target point to move to multiple preset positions. In order to solve the problem that the virtual stereo reference is not large to cover the whole measurement space, a new method is proposed. Firstly, by moving camera and capturing the reference images from multip direction, the entire measurement space can be covered by the virtual stereo references in all camera directions. Secondly, from each camera pose, a group of camera intrinsic and extrinsic parameters can be estimated. At last, by global optimizing the camera intrinsic parameters and the positons and orientations of the virtual stereo reference with respect to the camera, the optimal solution of calibration result can be obtained.4. Aim at the pose measurement of the hided part in the large-scale equipment, an assistant target with a number of IR-LEDs and a whole set of assistant target calibration method is proposed. Utilizing the algorithm of epipolar geometry and trigonometry to rebuild the spatial point coordinate, and using the accurate displacement of a single IR-LED controlled by movement equipment to recover the scale, the relative positions of target points on the assistant target can be acquired. To realize the indirect measurement, the relationship between target points coordinate frame and the measured part self coordinate frame needs to be calibrated. A “target/part” relative pose calibration method based on precisely controlled movement is presented, which using the relationship of both coordinate frames between multiple movement to establish the equations, the relative pose of them can be solved. At last, for the assistant target used for only attitude measurement, the “target/part” relative pose calibration is simplified, that significantly reduces the requirement of the equipment in the calibration method.5. A new target identity technology based on control of LED channels and the target distribution characteristic is proposed. This method solves the matching between image coordinates and spatial coordinates of targets. For the tracking problem of targets in pose measurement, a self-jointed tracing method combining the nearest-neighbor search and target distribution characteristic is proposed. This method keeps a quick search and when the tracking is failed caused by fast movement, set up connection with targets, and achieved the continuous tracking. Through the further study of perspective n pnints problem, using the orthogonal iteration method, the pose determination problem is solved.6. A prototype airborne equipment boresighting system(Archer-M) based on Monocular vision is designed and made with above key algorithms and techniques. The hardware platform is built and the overall framework, various function and the user interface of the software are achieved. All the algorithms proposed in this dissertation are all coded.The proposed algorithms and system functionalities have all been fully experimented and analyzed. It has been demonstrated that the prototype system has good measurement stability and high accuracy. The solution proposed in this paper has great potential to be widely used in real-time pose measurement for large-scale equipment assambly. |
PDF Full Text Request