The use of multiple cameras and geometric constraints for 3-D measurement
This thesis addresses some of the problems involved in the automation of 3-D photogrammetric measurement using multiple camera viewpoints. The primary research discussed in this thesis concerns the automatic solution of the correspondence problem. This and associated research has led to the development of an automated photograrnmetric measuring system which combines the techniques from both machine vision and photogrammetry. Such a system is likely to contribute greatly to the accessibility of 3-D measurement to non-photogrammetrists who will generally have little knowledge and expertise of photogrammetry. A matching method, which is called the 3-D matching method, is developed in the thesis. This method is based on a 3-D intersection and "epipolar plane", as opposed to the 2-D intersection of the epipolar line method. The method is shown to provide a robust and flexible procedure,especially where camera orientation parameters are not well known. The theory of the method is derived and discussed. It is further developed by combination with a bundle adjustment process to iteratively improve the estimated camera orientations and to gradually introduce legitimate matched target images from multiple cameras. The 3-D target matching method is also optimised using a 3-D space constrained search technique. A globally consistent search is developed in which pseudo target images are defined to overcome problems due to occlusion. Hypothesis based heuristic algorithms are developed to optimise the matching process. This method of solving target correspondences is thoroughly tested and evaluated by simulation and by its use in practical applications. The characteristics of the components necessary for a photogrammetric measuring system are investigated. These include sources of illumination, targets, sensors, lenses, and framegrabbers. Methods are introduced for analysis of their characteristics. CCD cameras are calibrated using both plumb line and self calibration methods. These methods provide an estimation of some of the sources of error, which influence the performance of the system as a whole. The design of an automated photogrammetric measuring system with a number of novel features is discussed and a prototype system is developed for use in a constrained environment. The precision, accuracy, reliability, speed, and flexibility of the developed system are explored in a number of laboratory and experimental applications. Trials show that with further development the system could have commercial value and be used to provide a solution suitable for photogrammetrists and trained operators in a wide range of applications.