3D-Environment Measurement System for Large-Scale Structure Manufacturing

This papcr presents a 3D-environment nieasurcmcnt systeni for large-scale structure manufacturing. First, a niethod and systeni for generating as-built data are shown. Then a camera calibration method that provides the ncces\ary nieasurenicnt precision in space is introduced, and the nieasurement error is analyzed quantitatively. Finally, the position and orientation of an ob,ject arc measured. and the experimental results are shown to support the theoretical conditions developed here. 1. lntrnduction For the asscnibling and maintaining of space structures, telcopcration of working niachine [8] will be researched for use in space until the late 1990's. In such situation, the telcoperation system niust know the manufacturing environnicnt. However, space structures are made of assenibled parts c.hich are manufactured with precision a few niillinicters. With such tolerances, errors in the space structures may occur easily, and installation of parts becomes difficult. This kind of problem is solved by measuring the nianufacturing environnient, storing the nieasured data in a database, and using this during the design process to ensure that the final design is well adapted to the nianufacturing environment. These dynamically changing data are called as-built data and systems that generate as-built data are called nianufacturing environnient nieasurement systems. Measurements are niade by using a measuring tape, a light wave surveying niachine. or a multilink structure. These nieasurenient methods can give the position of a point, but not the position and orientation of an object. Measurenients niade by using an image, however. can give the position and orientation of an object [ 7 ] . In space, it is necessary to measure objects that can not be touched. We therefore adopt the nieasurenient system using images in the manufacturing environnient nieasurenient systeni. Many researchers have worked on pattern detection, and we have taken particular notc of pattern niatching using a pattern field. Pattern detection in randoni fields by considering the diffusion field [ I ] has been reported, in which two parameters of the position of an object were estimated. The power and nionient in the potential fields were considered, and three parameters of the position and orientation of the object werc estimated [2]. Six-dcgree-offreedom contour niatching through single-eye noisy imagery was reported 161 and six paranieters of position and orientation of the ob,jcct werc estimated. In this papcr we adopt this algorithm for automatic pattcrn niatching. and we consider the nianufacturing environnient measurement systeni for nianufacturing large-scale structures. First, the niethod and the systeni generating the as-built data are shown. Then a camera calibration niethod that gives the necessary nieasurement precision in space is introduced, and the nieasurement error is analyzed quantitatively. Finally, the position and orientation of an ob.ject are measured, and the sample experiniental data is shown to support the theoretical conditions developed here. 2. Systeni Generating As-Ruilt Data During the manufacture of any particular object, the structure of the object is continually changing. Thus, its design is given concrete shape by the actual work process, for example, by manufacturing, or by carrying and installing structural parts. This changing qtructure niust be continually controlled by using design data. The precision of as-built data depends on the error of object shapes and on the installation error. The error of object shapes can be controlled during production since the production error is less than a few millimeters. This paper therefore considers the problem of installation error by establishing a measuring method. The structure must be nieasured to generate and keep asbuilt data. This, an interactive systeni [3] was constructed, using the design data as the prediction data for the position and orientation of the objects (Fig. 2.1). Fig. 2.1. Systcni Generating As-Built Data In this system, the design data is changed to as-built data by executing the following steps. ( I ) Object images are generated at the site by using CCD cameras, and are then displayed on workstations. (2) The position, orientation and shape data of the object is obtained froni the design database. (3) A wire-frame nlodel is graphically generated tionl the design data, and is indicated in the same place as the image froni the caniera. (4) An operator roughly matches the wire-frame model to the iniage niodel on the display by anlending the position and orientation of the object. Also, thc wire-frame model is matched to the iniage niodel by using the recursive algorithm of six-degree-of-freedom contour matching. ( 5 ) The position and orientation of the object are saved in the database. The changing structure can be controlled by iterating the above method. The object position based on the caniera position is determined by matching the wire-frame model to the image model. The measurement precision of the position and orientation of the object is assured by an automatic measuring algorithm [S]. The object position based on the coordinates fixed in the environment must be determined in order to generate the as-built data. Therefore, an appropriate method is to use a surveying niachine with a camera, such as that shown in Fig. 2.2. The position and orientation of the object is measured by combining the image with the position and orientation data of the camera. Fig. 2.2. Surveying Machine with Camera Although many parts are used in large-scale structure construction, few types of parts are used. Thus, the asbuilt data structure shown in Fig. 2.3 is appropriate. Fig. 2.3. As-Built Data Structure In Fig. 7.3, each symbol represents a particular type of ob.ject. There are various types, such as plates, sleeves, pipes, and so on. The class represents each ob.ject shape. For example, the shape of a plate is represented by the lengths of its side and its height, the shape of a sleeve is represented by its diameter, and the shape of a pipe is represented by its diameter and its length. The instance represents the position and orientation of the object. 3. Image Process Identification 3.1 lnlage Process hlodel In the as-built data generation system shown in Fig. 2 . I, the iniage of the structure is generated by using the surveying machine and camera (Fig. ? . 2 ) , and the image is displayed on the workstation. The position and orientation of the camera are obtained roughly from the position and orientation of the surveying machine nieawred by using the surveying machine. The image process is therefore represented by the following equation: