Reverse engineering requires the acquisition of large amounts of data describing the surface of an object, sufficient to replicate that object accurately using appropriate fabrication techniques. This is important within a wide range of commercial and scientific fields where CAD models may be unavailable for parts that must be duplicated or modified, or where a physical model is used as a prototype. The three-dimensional digitisation of objects is an essential first step in reverse engineering. Optical triangulation laser sensors are one of the most popular and common non-contact methods used in the data acquisition process today. They provide the means for high resolution scanning of complex objects. Multiple scans of the object are usually required to capture the full 3D profile of the object. A number of factors, including scan resolution, system optics and the precision of the mechanical parts comprising the system may affect the accuracy of the process. A single perspective optical triangulation sensor provides an inexpensive method for the acquisition of 3D range image data. However, there are often locations within each scan where data is seriously flawed because the data acquisition process is subject to distortions. Such distortions are often associated with edges in the object, where regions of high curvature (relative to the incident angle of the sensor) cause occlusions and secondary reflections of the laser beam, resulting in false height readings. Abrupt changes in surface reflectance or texture can also have similar effects. Previous work has determined that the orientation of the scan head with respect to the edges of the object is a major factor in the degree of such distortions. Combining multiple range images using compensation algorithms has reduced the level of distortion in the integrated data set; however capturing the number of necessary repetitions of the entire scan is very time-consuming. A development platform has been established to investigate how data distortions may be reduced by the application of image analysis techniques in planning the scan process. By using information on edge location and orientation recovered from a digital camera image, partial scans may then be performed for each determined orientation of the scanner, thereby avoiding much redundant coverage of the entire scan area. Vectorisation algorithms, based on known edge detection techniques, have been developed to determine the position of vectors corresponding to the discovered edges. Further algorithms have been developed to process these vectors into 'scan regions' corresponding to each particular scanner orientation. When the object is scanned at the orientation corresponding to the scan region the distortions are likely to be much reduced. Some features of the object geometry, such as small holes or internal corners present a particular problem where a number of scan regions representing different scan orientations overlap. Because of the nature of the scanner such regions are liable to show some level of distortion for all laser orientations. However, these locations can be identified from the camera image and the user alerted to the presence of unreliable data. Calibration methods relating the image and scan space have been shown to be susceptible to errors caused by optical effects from the camera, such as lens barrel distortion and errors due to parallax. Algorithms have been developed to compensate for these effects and combine the data from a number of partial scans in order to provide a single integrated point cloud.
| Identifer | oai:union.ndltd.org:bl.uk/oai:ethos.bl.uk:442098 |
| Date | January 2006 |
| Creators | Denby, Alistair John |
| Publisher | Nottingham Trent University |
| Source Sets | Ethos UK |
| Detected Language | English |
| Type | Electronic Thesis or Dissertation |
| Source | http://irep.ntu.ac.uk/id/eprint/210/ |
Page generated in 0.0022 seconds