Optical three-dimensional (3D) measurement technologies have significant contributions to many industries such as manufacturing, security and entertainment. A binocular vision system with structured-light has the highest accuracy for the inspection tasks that require large working space, high measurement speed and high scanning resolution. However, its accuracy still cannot fulfill the demands from the advanced manufacturing and the 3C (computer, communications, consumer electronics) industries which fall into the range of several micrometers. The accuracy of 3D measurement cannot be higher, because the resolution of the imaging sensor cannot be fully utilized and defocusing error is induced due to limited depth of field (DOF) at high magnification. Both problems are caused by the pin-h...[ Read more ]

Optical three-dimensional (3D) measurement technologies have significant contributions to many industries such as manufacturing, security and entertainment. A binocular vision system with structured-light has the highest accuracy for the inspection tasks that require large working space, high measurement speed and high scanning resolution. However, its accuracy still cannot fulfill the demands from the advanced manufacturing and the 3C (computer, communications, consumer electronics) industries which fall into the range of several micrometers. The accuracy of 3D measurement cannot be higher, because the resolution of the imaging sensor cannot be fully utilized and defocusing error is induced due to limited depth of field (DOF) at high magnification. Both problems are caused by the pin-hole model.

To solve the above two problems, the thin-lens model is adopted, instead of the pin-hole model. Accurate calibration and measurement at high magnification are available with the thin-lens model. Meanwhile, the thin-lens model can ensure the scene is in focus at all times by following the Gaussian optical formula, which extends the DOF of measurement systems.

To preliminarily demonstrate the superiority of fully utilizing the resolution of the imaging sensor, a differential measurement method based on the thin-lens model is proposed to accurately measure the paraxial focal length of a lens with a simple apparatus.

To facilitate the development of the thin-lens model in machine vision applications, two types of thin-lens cameras are studied. One is the camera system with movable imaging sensor (MIS). The other is the camera system with electrically tunable lens (ETL) group based on shape-changing polymer (SCP). A unified thin-lens camera model of these thin-lens camera systems for machine vision applications is established based on detailed analyses of their optical properties. The corresponding calibration methods, including initial estimation of camera parameters and bundle adjustment (BA) strategy, are respectively developed for two types of thin-lens camera systems. A real time temperature compensation method is also developed for the thin-lens camera system with ETL group. Both simulations and experiments were conducted to verify the effectiveness and accuracy of the proposed calibration methods. With the proposed calibration methods for the thin-lens camera systems, the calibration workload is significantly reduced and accurate calibration at high magnification is achieved. Meanwhile, the potential of applying thin-lens camera systems in machine vision applications is unleashed.

To fulfill the high-accuracy requirement from the industries, a binocular vision system with structured-light based on the thin-lens model (BVSSTM) is developed in this thesis. The system model, stereo calibration strategy, bundle adjustment strategy, correspondence matches and measurement principle are presented in the research. With BVSSTM, accurate measurement with extended DOF for small objects can be achieved. This unleashes the potential of fully automating the inspection process in the advanced manufacturing and 3C industries. All these significances are benefited from the theories of the thin-lens model. Once the thin-lens model is utilized, it can spark further impact in other vision-based applications.