Significance  Transparent objects such as glass, lenses, and displays have a large number of needs and applications in different fields, which need to be reconstructed in 3D (three-dimensional) quickly, accurately and completely, so as to objectively describe the object’s profile, inspect the surface quality and ensure the function of the components. However, due to the specular reflection and refraction characteristics of transparent surfaces, the traditional 3D reconstruction methods can’t be directly applied to 3D topography measurements. This dissertation focuses on the non-intrusive 3D reconstruction methods for transparent objects, and reviews the main research work in recent years from two aspects: reconstruction based on reflected light and reconstruction based on transmitted light. Subsequently, the advantages, disadvantages, and applicable scenarios of various techniques are detailed. Users can select appropriate reconstruction schemes for transparent objects based on different application requirements and measurement conditions. At the same time, it also looks forward to the future research direction of this field, expecting to provide ideas for researchers to improve existing methods or explore new ones.

Progress  Most of the various non-invasive measurement systems and technologies use cameras to capture deformed patterns that have been reflected or refracted on the surface of transparent objects to restore 3D topography. These patterns carry a wealth of information about the surface shape of the transparent object, but they also contain a lot of detail in the surrounding background caused by the reflection and refraction of light, i.e., this distortion is the result of the combination of the surface shape, the surrounding background, and the lighting conditions. Researchers have done a lot of work to extract the appearance of transparent objects from it. Depending on the type of optical transmission mode in the measurement, this dissertation divides it into reflected light-based reconstruction methods and transmitted light-based reconstruction methods.

Reflected light-based reconstruction methods include scan-based methods, polarization and its combination with other techniques, and reflective phase measuring deflectometry (RPMD). Transmitted light-based reconstruction methods include: shape from distortion, transmission phase measuring deflectometry (TPMD), stereo vision, etc. The principles and current development status of each method are introduced, along with a summary of their application scenarios, key advantages, and disadvantages. To provide a comprehensive evaluation, these methods are further compared from multiple perspectives, including speed, accuracy, cost, computational complexity, and calibration complexity.

Finally, this dissertation summarizes the general situation of non-intrusive 3D reconstruction methods for transparent objects, and looks forward to the possible future development directions.

Conclusions and Prospects  This dissertation aims to provide an updated and general overview of non-intrusive 3D reconstruction methods for transparent objects. The measurement speed of the scanning method is limited by the way it is scanned on a point-by-point or line-by-line basis. Shape recovery from polarization is characterized by fast detection speed, high accuracy, and easy calibration. The accuracy and speed of RPMD depend on the specific method used to deal with parasitic reflections. TPMD is competitive in wavefront measurement and defect detection. Stereo vision technology, with its high accuracy and low cost, has become a powerful tool for dealing with complex transparent objects. Deep learning and multimodal fusion have demonstrated new vitality in the 3D reconstruction of transparent objects, and future developments are expected to increase speed, reduce costs, and enhance environmental adaptability and robustness.